This PDF file contains the front matter associated with SPIE Proceedings Volume 10420 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

Age is an inherent biological characteristic of human and is reflected in facial images to a certain extent. A method for estimating age from a facial image by combining CNN (Convolution Neural Network) with SVR (Support Vector Regression) is proposed. First, a deep CNN is trained to automatically extract age features from facial images and classify them into variant age groups. Then different SVRs are trained for each age group to estimate the age of a facial image. The experimental results show that a lower MAE (Mean Absolute Error) of age estimation on MORPH database is obtained.

The performance and robustness of fatigue detection largely decrease if the driver with glasses. To address this issue, this paper proposes a practical driver fatigue detection method based on face alignment at 3000 FPS algorithm. Firstly, the eye regions of the driver are localized by exploiting 6 landmarks surrounding each eye. Secondly, the HOG features of the extracted eye regions are calculated and put into SVM classifier to recognize the eye state. Finally, the value of PERCLOS is calculated to determine whether the driver is drowsy or not. An alarm will be generated if the eye is closed for a specified period of time. The accuracy and real-time on testing videos with different drivers demonstrate that the proposed algorithm is robust and obtain better accuracy for driver fatigue detection compared with some previous method.

Compensating the illumination of a face image is an important process to achieve effective face recognition under severe illumination conditions. This paper present a novel illumination normalization method which specifically considers removing the illumination boundaries as well as reducing the regional illumination. We begin with the analysis of the commonly used reflectance model and then expatiate the hybrid usage of adaptive non-local smoothing and the local information coding based on Weber’s law. The effectiveness and advantages of this combination are evidenced visually and experimentally. Results on Extended YaleB database show its better performance than several other famous methods.

Based on the maximum margin criterion (MMC), a new algorithm of statistically uncorrelated optimal discriminant vectors and a new algorithm of orthogonal optimal discriminant vectors for feature extraction were proposed. The purpose of the maximum margin criterion is to maximize the inter-class scatter while simultaneously minimizing the intra-class scatter after the projection. Compared with original MMC method and principal component analysis (PCA) method, the proposed methods are better in terms of reducing or eliminating the statistically correlation between features and improving recognition rate. The experiment results on Olivetti Research Laboratory (ORL) face database shows that the new feature extraction method of statistically uncorrelated maximum margin criterion (SUMMC) are better in terms of recognition rate and stability. Besides, the relations between maximum margin criterion and Fisher criterion for feature extraction were revealed.

The Web application presented here supports plant production and works with the graph database Neo4j shell to support the assessment of the condition of crops on the basis of geospatial data, including raster and vector data. The adoption of a graph database as a tool to store and manage the data, including geospatial data, is completely justified in the case of those agricultural holdings that have a wide range of types and sizes of crops. In addition, the authors tested the option of using the technology of Microsoft Cognitive Services at the level of produced application that enables an image analysis using the services provided. The presented application was designed using ASP.NET MVC technology and a wide range of leading IT tools.

In the universal computerization era important aspect is the possibility of man partial replacement in the identification process of the disease cases occurring in the plantations cultivation. This is very important when it comes to the process of early response to the disease and at a later stage for early treatment to help stop the disease or its elimination, which has its economic justification. Presented in this paper system may be the answer to these issues in the production of sugar beet

In order to deal with the insufficiency of recently algorithms based on Two Dimensions Fractional Fourier Transform (2D-FrFT), this paper proposes a multiple order features based method for emotion recognition. Most existing methods utilize the feature of single order or a couple of orders of 2D-FrFT. However, different orders of 2D-FrFT have different contributions on the feature extraction of emotion recognition. Combination of these features can enhance the performance of an emotion recognition system. The proposed approach obtains numerous features that extracted in different orders of 2D-FrFT in the directions of x-axis and y-axis, and uses the statistical magnitudes as the final feature vectors for recognition. The Support Vector Machine (SVM) is utilized for the classification and RML Emotion database and Cohn-Kanade (CK) database are used for the experiment. The experimental results demonstrate the effectiveness of the proposed method.

Human gait, as a soft biometric, helps to recognize people by walking. To further improve the recognition performance under cross-view condition, we propose Joint Bayesian to model the view variance. We evaluated our prosed method with the largest population (OULP) dataset which makes our result reliable in a statically way. As a result, we confirmed our proposed method significantly outperformed state-of-the-art approaches for both identification and verification tasks. Finally, sensitivity analysis on the number of training subjects was conducted, we find Joint Bayesian could achieve competitive results even with a small subset of training subjects (100 subjects). For further comparison, experimental results, learning models, and test codes are available.

In complex scene, the exact determination and division of color is the key to color flag recognition tasks. Aiming at the problem that the traditional recognition methods have low recognition accuracy and can not be identified in complex scene and multi-targets, this paper proposes a color threshold determination (CTD) method to identify color flag. Firstly, the color sample data set under different light conditions is constructed, and the filtering of the noise pixels is realized by the projection method in the HSV color space to obtain the final color decision threshold. Secondly, color decision is made for each pixel in the candidate region of interest detected by Gentle Adaboost cascade classifier based on HOG feature. Finally, the color feature is matched by the preset threshold, and the matching region is retained to obtain the final recognition results. Experiments show that the proposed approach has better performance in the color flag recognition task under complex scene. The recognition accuracy was 97.1%, the sensitivity was 90.70% and the specificity was 99.33%.

In this paper, we present a technique to automatically identify some typical radio frequency interference from pulsar surveys using support vector machine. The technique has been tested by candidates. In these experiments, to get features of SVM, we use principal component analysis for mosaic plots and its classification accuracy is 96.9%; while we use mathematical morphology operation for smog plots and horizontal stripes plots and its classification accuracy is 86%. The technique is simple, high accurate and useful.

The reconstruction of 3D objects from 2D line drawings is regarded as one of the key topics in the field of computer vision. The ongoing research is mainly focusing on the reconstruction of 3D objects that are mapped only from 2D straight lines, and that are symmetric in nature. Commonly, this approach only produces basic and simple shapes that are mostly flat or rather polygonized in nature, which is normally attributed to inability to handle curves. To overcome the above-mentioned limitations, a technique capable of handling non-symmetric drawings that encompass curves is considered. This paper discusses a novel technique that can be used to reconstruct 3D objects containing curved lines. In addition, it highlights an application that has been developed in accordance with the suggested technique that can convert a freehand sketch to a 3D shape using a mobile phone.

Recently, multimodal biometric systems have received considerable research interest in many applications especially in the fields of security. Multimodal systems can increase the resistance to spoof attacks, provide more details and flexibility, and lead to better performance and lower error rate. In this paper, we present a multimodal biometric system based on face and ear, and propose how to exploit the extracted deep features from Convolutional Neural Networks (CNNs) on the face and ear images to introduce more powerful discriminative features and robust representation ability for them. First, the deep features for face and ear images are extracted based on VGG-M Net. Second, the extracted deep features are fused by using a traditional concatenation and a Discriminant Correlation Analysis (DCA) algorithm. Third, multiclass support vector machine is adopted for matching and classification. The experimental results show that the proposed multimodal system based on deep features is efficient and achieves a promising recognition rate up to 100 % by using face and ear. In addition, the results indicate that the fusion based on DCA is superior to traditional fusion.

The order of cigarette market is a key issue in the tobacco business system. The anti-counterfeiting code, as a kind of effective anti-counterfeiting technology, can identify counterfeit goods, and effectively maintain the normal order of market and consumers' rights and interests. There are complex backgrounds, light interference and other problems in the anti-counterfeiting code images obtained by the tobacco recognizer. To solve these problems, the paper proposes a locating method based on Susan operator, combined with sliding window and line scanning,. In order to reduce the interference of background and noise, we extract the red component of the image and convert the color image into gray image. For the confusing characters, recognition results correction based on the template matching method has been adopted to improve the recognition rate. In this method, the anti-counterfeiting code can be located and recognized correctly in the image with complex background. The experiment results show the effectiveness and feasibility of the approach.

The goal of sign language recognition (SLR) is to translate the sign language into text, and provide a convenient tool for the communication between the deaf-mute and the ordinary. In this paper, we formulate an appropriate model based on convolutional neural network (CNN) combined with Long Short-Term Memory (LSTM) network, in order to accomplish the continuous recognition work. With the strong ability of CNN, the information of pictures captured from Chinese sign language (CSL) videos can be learned and transformed into vector. Since the video can be regarded as an ordered sequence of frames, LSTM model is employed to connect with the fully-connected layer of CNN. As a recurrent neural network (RNN), it is suitable for sequence learning tasks with the capability of recognizing patterns defined by temporal distance. Compared with traditional RNN, LSTM has performed better on storing and accessing information. We evaluate this method on our self-built dataset including 40 daily vocabularies. The experimental results show that the recognition method with CNN-LSTM can achieve a high recognition rate with small training sets, which will meet the needs of real-time SLR system.

Flag is a rather special recognition target in image recognition because of its non-rigid features with the location, scale and rotation characteristics. The location change can be handled well by the depth learning algorithm Convolutional Neural Networks (CNNs), but the scale and rotation changes are quite a challenge for CNNs. Since it has good rotation and gray scale invariance, the local binary pattern (LBP) is combined with grayscale stretching and CNNs to make LBP and grayscale stretching as CNNs pretreatment, which can not only significantly improve the efficiency of flag recognition, but can also evaluate the recognition effect through ROC, accuracy, MSE and quality factor.

With the progress of smartphones hardware, it is simple on smartphone using image recognition technique such as face detection. In addition, indoor navigation system development is much slower than outdoor navigation system. Hence, this research proves a usage of image recognition technique for navigation in indoor environment. In this paper, we introduced an indoor navigation application that uses the indoor environment features to locate user’s location and a route calculating algorithm to generate an appropriate path for user. The application is implemented on Android smartphone rather than iPhone. Yet, the application design can also be applied on iOS because the design is implemented without using special features only for Android. We found that digital navigation system provides better and clearer location information than paper map. Also, the indoor environment is ideal for Image recognition processing. Hence, the results motivate us to design an indoor navigation system using image recognition.

In this paper, we propose a novel approach based on Depth Maps and compute Multi-Scale Histograms of Oriented Gradient (MSHOG) from sequences of depth maps to recognize actions. Each depth frame in a depth video sequence is projected onto three orthogonal Cartesian planes. Under each projection view, the absolute difference between two consecutive projected maps is accumulated through a depth video sequence to form a Depth Map, which is called Depth Motion Trail Images (DMTI). The MSHOG is then computed from the Depth Maps for the representation of an action. In addition, we apply L2-Regularized Collaborative Representation (L2-CRC) to classify actions. We evaluate the proposed approach on MSR Action3D dataset and MSRGesture3D dataset. Promising experimental result demonstrates the effectiveness of our proposed method.

In order to solve the problem of the subtle inter-class difference and large intra-class variation caused by different poses in fine grained object recognition, a new algorithm is proposed for fine-grained object recognition based on pose alignment and part based representation. In the training stage, firstly, dozens of typical poses of the object are summed up by clustering, and then a pose classifier is trained based on them which can be used to align object pose. Then, the fine-grained classifiers are trained by the visible parts and global feature ensembles in each pose subclass. Given an unseen test image, the probabilities that current test image belongs to each pose subclass are gotten via the pose classifier. The output from first k subclass classifiers is weighted combined to form a single classification decision. Our approach is especially valuable for fine-grained recognition when intra-class variations are extremely high. The proposed method is validated by the experimental results on the bird database CUB_2010-2011, which has the large intra-class variations. The results show that the proposed method has better performance than the existing methods.

Multi-object tracking is particularly challenging in many scenarios with similar appearance and frequent occlusions among targets. In this paper, we present an online detection-based multi-object tracking method. In each frame, kernerlized convolution filter are adopted to track isolated and un-occluded targets. To overcoming fixed scale in KCF, trackers are associated with detection responses. If a target is associated with a detection, then the target size is updated by the average of this detection size and the previous estimated size. When occlusions are detected, the multiple interaction among targets is formulated as an optimization problem and we explore two-layer hierarchical Particle Swarm Optimization algorithm for the optimal solution. The first layer is designed for the superficial targets which is visible. The second layer is designed for the bottom occluded targets which can guided by first visible layer and we propose to incorporate the attractive force into the particle evolution process. Experimental results on public datasets demonstrate that our proposed method alleviating drifting problem and effectively reduces ID switches and lost trajectories.

Multi-object tracking in semi-crowded environment is a difficult task. Traditional methods relying only on visual cues, like appearance and simple motion prediction fails because of frequent occlusions. In this paper, a new method based on formation stability is proposed to learn the interaction information among pedestrians. whatever the relationship of a pedestrian with others is, it will remain unchanged for a while. That is, when a pedestrian is occluded, the relationship with others will stay the same after the reappearance of the pedestrian. So trajectory clips caused by occlusions can be associated together. Effectiveness of this method is validated by experiments.

In this paper, a novel target tracking algorithm based on spatio-temporal context is proposed. During the tracking process, the camera shaking or occlusion may lead to the failure of tracking. The proposed algorithm can solve this problem effectively. The method use the spatio-temporal context algorithm as the main research object. We get the first frame’s target region via mouse. Then the spatio-temporal context algorithm is used to get the tracking targets of the sequence of frames. During this process a similarity measure function based on perceptual hash algorithm is used to judge the tracking results. If tracking failed, reset the initial value of Mean Shift algorithm for the subsequent target tracking. Experiment results show that the proposed algorithm can achieve real-time and stable tracking when camera shaking or target occlusion.

In order to achieve real-time target tracking and locating for UAV, a reliable processing system is built on the embedded platform. Firstly, the video image is acquired in real time by the photovoltaic system on the UAV. When the target information is known, KCF tracking algorithm is adopted to track the target. Then, the servo is controlled to rotate with the target, when the target is in the center of the image, the laser ranging module is opened to obtain the distance between the UAV and the target. Finally, to combine with UAV flight parameters obtained by BeiDou navigation system, through the target location algorithm to calculate the geodetic coordinates of the target. The results show that the system is stable for real-time tracking of targets and positioning.

Traditional wavelet threshold denoising algorithm is insufficient for the preservation of edge detail information and the separation of noise. In this paper, proposed the optimization of wavelet threshold denoising method based on edge detection, which to process the edge image obtained by the wavelet edge detection algorithm, and fuse the smoothing image produced by the improved threshold function model. Experimental results show that compared with the traditional wavelet threshold denoising method, this algorithm effectively preserves the edge information of the image and remove the noise, also improved signal to noise ratio obviously

A surface defect detection method based on multi-features is proposed for the detection of complex ceramic disc surface. Firstly, geometric transformation is implemented on the image being detected, which makes the direction and position of the image to be unified. Then, according to the standard template, the image is separated into several regions with different gray characteristics for interval inspect. Defects containing broken, chipping, uneven polishing, crack, etc. are classified into two types—blob defect and crack defect. Blob defect detection is mainly based on the analysis of the abnormal part of regional gray value, and the crack detection is based on the analysis of shape and neighborhood gray gradient changes. Experimental result shows that the algorithm proposed is sensitive to the surface defects, and can effectively reduce the false positives with detection rate more than 94%, detection speed up to 5 parts per second.

In response to the slow running speed of the Deformable Part Model algorithm in the process of the pedestrian detection, this paper incorporated Cascade Detection algorithm and Branch-and-Bound algorithm into a fast pedestrian detection algorithm which is based on Deformable Part Model. In a pedestrian detection process, a sequence model evaluates individual parts sequentially to quickly prune most of the smaller possible objects. This aims to accelerate the process of object positioning, and to optimize global classification results in all possible image regions. Meanwhile, the boundaries of the maximum are adopted to search the clipping operation of the window. In order to improve detection speed without compromising the accuracy of the detection, this paper increase the number of the part models involved. According to the experimental results on INRIA data set, the proposed algorithm successfully improved the accuracy and the speed of detection.

Object detection is a computer vision problem which caught a large amount of attention. But the candidate boundingboxes extracted from only image features may end up with false-detection due to the semantic gap between the top-down and the bottom up information. In this paper, we propose a novel method for generating object bounding-boxes proposals using the combination of eye fixation point, saliency detection and edges. The new method obtains a fixation orientated Gaussian map, optimizes the map through single-layer cellular automata, and derives bounding-boxes from the optimized map on three levels. Then we score the boxes by combining all the information above, and choose the box with the highest score to be the final box. We perform an evaluation of our method by comparing with previous state-ofthe art approaches on the challenging POET datasets, the images of which are chosen from PASCAL VOC 2012. Our method outperforms them on small scale objects while comparable to them in general.

Automatic inspection takes a great role in guaranteeing the product quality. But one of the limitations of current inspection algorithms is either product specific or problem specific. In this paper, we propose a defect detection method based on three image features fusion for variety of industrial products surface detection. The proposed method learns sub-image gray level difference, color histogram and pixel regularity of qualified images off-line and test the images based on the detection results of these three image features. It avoids the feature training of defect products as it is difficult to collect large amount of defect samples. The experimental results show that the detection accuracy is between 93% and 98% and the approach is efficient for the real time applications of industrial product inspect.

In this paper, we propose a novel cascade detection algorithm which focuses on point and line defects on TFT-LCD. At the first step of the algorithm, we use the gray level difference of su-bimage to segment the abnormal area. The second step is based on phase only transform (POT) which corresponds to the Discrete Fourier Transform (DFT), normalized by the magnitude. It can remove regularities like texture and noise. After that, we improve the method of setting regions of interest (ROI) with the method of edge segmentation and polar transformation. The algorithm has outstanding performance in both computation speed and accuracy. It can solve most of the defect detections including dark point, light point, dark line, etc.

The change point detection in periodic time series is much desirable in many practical usages. We present a novel algorithm for this task, which includes two phases: 1) anomaly measure- on the basis of a typical regression model, we propose a new computation method to measure anomalies in time series which does not require any reference data from other measurement(s); 2) change detection- we introduce a new martingale test for detection which can be operated in an unsupervised and nonparametric way. We have conducted extensive experiments to systematically test our algorithm. The results make us believe that our algorithm can be directly applicable in many real-world change-point-detection applications.

Indoor Autonomous Vehicle(IAV) is used in many indoor scenes. Such as hotels and hospitals. Door detection is a key issue to guide the IAV into rooms. In this paper, we consider door detection in the use of indoor navigation of IAV. Since real-time properties are important for real-world IAV, the detection algorithm must be fast enough. Most monocular-camera based door detection model need a perfect detection of the four line segments of the door or the four corners. But in many situations, line segments could be extended or cut off. And there could be many false detected corners. And few of them can distinguish doors from door-like objects with door-like shape effectively. We proposed a 2-D vision model of the door that is made up of line segments. The number of parts detected is used to determine the possibility of a door. Our algorithm is tested on a database of doors.1 The robustness and real-time are verified. The precision is 89.4%. Average time consumed for processing a 640x320 figure is 44.73ms.

This paper introduces a joint feature of Fourier histograms of oriented gradients (FHOG) and local binary pattern (LBP) for vehicle detection in aerial images. Both of them are rotation invariant, so any rotation angle of vehicle in aerial images can be easily detected. A linear support vector machine (SVM) classifier is then trained over the joint feature vectors for the final vehicle detection. We evaluate our method on a public dataset and compare with some state-of-theart algorithms, the proposed joint feature outperforms them in detecting small targets in complicated backgrounds.

Infrared small target detection is one of the vital techniques in infrared search and track surveillance systems. An efficient method based on target-background separation via local morphological component analysis (MCA) sparse representation is proposed in this paper. This method converts infrared small target detection problem over entire image into target-background separation over image patches according to the different morphological component between target and background. An adaptive dictionary is trained adaptively by K-singular value decomposition (K-SVD) according to infrared image, and then the dictionary is subdivided by a total-variation-like activity measure into two categories: the target component dictionary explaining target signal and background component dictionary embedding background. Finally, the interest target can be easily extracted through threshold segmentation in target component image constructed by target dictionary. The experimental results demonstrate the effectiveness of the proposed method.

Offshore accurate replenishment technology is the foundation for conducting ocean research for every country. However, it is difficult to keep the security and accuracy of hoisting due to the fact that waves will let vessels generate 6-DOF motions. This paper regards accurate offshore supplying as a background, and takes into consideration that vessel acts as a rigid body to perform algorithm research on the AHC (Active Heaving Compensation) detection system. The binocular camera installed on the hoisting equipment can calculate the 6-DOF pose of vessel via detecting landmarks on the deck. The system can achieve all-weather operations, adopting Shi-Tomasi algorithm to identify and L-K Optical Flow algorithm to track sub-pixel points. Lastly, the scheme has been verified in the 6-DOF motion platform, which indicates that its accuracy meets the requirements of the control experiment in the next step.

A vehicle detection and tracking system is one of the indispensable methods to reduce the occurrence of traffic accidents. The nearest vehicle is the most likely to cause harm to us. So, this paper will do more research on about the nearest vehicle in the region of interest (ROI). For this system, high accuracy, real-time and intelligence are the basic requirement. In this paper, we set up a system that combines the advanced KCF tracking algorithm with the HaarAdaBoost detection algorithm. The KCF algorithm reduces computation time and increase the speed through the cyclic shift and diagonalization. This algorithm satisfies the real-time requirement. At the same time, Haar features also have the same advantage of simple operation and high speed for detection. The combination of this two algorithm contribute to an obvious improvement of the system running rate comparing with previous works. The detection result of the HaarAdaBoost classifier provides the initial value for the KCF algorithm. This fact optimizes KCF algorithm flaws that manual car marking in the initial phase, which is more scientific and more intelligent. Haar detection and KCF tracking with Histogram of Oriented Gradient (HOG) ensures the accuracy of the system. We evaluate the performance of framework on dataset that were self-collected. The experimental results demonstrate that the proposed method is robust and real-time. The algorithm can effectively adapt to illumination variation, even in the night it can meet the detection and tracking requirements, which is an improvement compared with the previous work.

Considering that traditional image edge detection methods can’t reflect the orientation selection characteristics of a human visual perception system and the firing mechanism of neuron spikes, a new method of image edge detection based on orientation response mechanisms of an integrate-and-fire (IF) neurons model are presented in this paper. To fully reflect orientation selection characteristics of the human visual system, the Log-Gabor orientation response model is applied for image orientation preprocessing. The spikes sequence fired by IF neurons model is used for image edge detection. The results of various experiments show that our method can truly reflect the biological characteristics. By comparison with the traditional image edge detection methods, our method is focused on the fine details preservation, and can highlight image edge.

Random Pulse Repetition Interval Compressed Sensing Radar (RPRICSR) has superiorities on target detection, unambiguous velocity measurement as well as anti-velocity jamming for its signal randomness. However, RPRICSR cannot detect the micro motion targets effectively under the presence of micro motion false targets within present echo processing methods. This paper firstly combined Short Sparse Recovery (SSR) method with RPRI signal under the frame of Compressed Sensing. Then the fact that echo signals of true and false micro motion targets are sparse in different dictionaries are utilized to separate the true and false micro motion targets based on a union dictionary. The proposed method is proved to be effective compared to the traditional signal processing methods of RPRICSR according to the simulation results.

For intelligent vehicle surveillance systems, it is a big challenge to detect small, blurred car plates of vehicles driving on a highway. In this paper, we present a novel, two-stage detection scheme for small, blurred car-plate detection in large surveillance images. Our proposed scheme firstly detects vehicles, and then locates the car plates in specific regions of detected vehicles based on our proposed car-face landmark localization algorithm. Our scheme can also solve the high false-alarming rate problem with small, blurred car-plate detection. Experimental results show that our proposed method is accurate, and able to reduce the false-alarming rate, without any compromise in speed.

Moving target detection in image sequence for dynamic scene is an important research topic in the field of computer vision. Block projection and matching are utilized for global motion estimation. Then, the background image is compensated applying the estimated motion parameters so as to stabilize the image sequence. Consequently, background subtraction is employed in the stabilized image sequence to extract moving targets. Finally, divide the difference image into uniform grids and average optical flow is employed for motion analysis. Experiment tests show that the proposed average optical flow method can efficiently extract the vehicle targets from dynamic scene meanwhile decreasing the false alarm.

Gaussian Mixture Model is often employed to build background model in background difference methods for moving target detection. This paper puts forward an adaptive moving target detection algorithm based on improved Gaussian Mixture Model. According to the graylevel convergence for each pixel, adaptively choose the number of Gaussian distribution to learn and update background model. Morphological reconstruction method is adopted to eliminate the shadow.. Experiment proved that the proposed method not only has good robustness and detection effect, but also has good adaptability. Even for the special cases when the grayscale changes greatly and so on, the proposed method can also make outstanding performance.

Structured light 3D vision inspection is widely used in various 3D surface reconstruction techniques. However, there are still two problems to be solved. The first problem is that if the relative position between the camera and laser projector is changed, the system has to be calibrated again. The second problem is that a turntable or a linear positioning system is needed in order to obtain the complete 3D reconstruction data in conventional methods. To overcome these problems, we proposed a novel design using a dual semicircular planar target and a rotating laser projector to perform calibration and 3D reconstruction simultaneously. The system contains two semicircular planar targets hollowed out at center which are installed in up and down position. The middle part of the light stripe can pass through the blank part of semicircular planar target and then cast on the target object, while the rest of the light stripe will cast on the semicircular planar target. The laser projector is rotated 360 degrees along an axis perpendicular to the planar target plane to get the complete 3D surface reconstruction of the target object. Experiments are conducted to validate the flexibility and accuracy of the system.

Human vision system possesses a strong ability to interpret complex scenes in real time effortlessly and efficiently. However, computational modeling of basic intelligent behavior still remains a significant challenge in the fields of computer vision system. Different from context-aware saliency algorithm, our method for saliency detection, not only using basic properties of visual attention but also introducing local patches contrast cues, thus it can effectively improve the accuracy of visual saliency detection with relatively low computational complexity. Furthermore, the experimental results also indicate that the effectiveness of our proposed method.

In order to achieve a single-frame enhancement and detection of dim point targets efficiently, a new method based on morphological top-hat transformation and modified non-local means (NL-means, NLM) for target enhancement is presented. After enhancing dim point targets, an estimating algorithm called local reverse entropy is applied to get candidate targets, and obtains detecting results in the end. In this model, white top-hat and black top-hat are combined to obtain pre-processed image, then the targets are enhanced for the second time through modified NL-means algorithm. The background in the residual image of enhanced image and original image is mostly suppressed, then it is regarded as the input of local reverse entropy estimation method (LREM). Detecting results can be obtained by setting proper thresholds. The 4 × 4 weak lattice targets with different brightness are superimposed on different infrared image backgrounds. The experimental results show that, when the SCR is low (SCR≈1), the detection algorithm model proposed in this paper has higher SCR gain than other target-enhancing algorithms such as TDLMS, max-median, max-mean, non-local means, etc. and the detection performance is the best.

In this paper, we propose a novel approach to detect visual saliency based on improved graph models. The graph-based manifold ranking algorithm adopts k neighbor constraint and close-loop constraint to construct a graph model. In the first stage of the algorithm, k = 2 neighbor constraint reduces the differences between foreground nodes and boundary nodes and the saliency value of the nodes which are close to the center of the image tends to be larger. So some background nodes are selected as foreground queries in the second stage and some background will be highlighted. To eliminate such defect of the algorithm, we use a simple graph structure containing close-loop constraint in the first stage and update the graph structure by expanding close-loop constraint to calculate a new ranking matrix, which is then used to calculate the final saliency map. Experiments in the MSRA1000 database prove that our method outperforms the classical algorithm.

Since the study of fuel consumptions is of great importance and the related data is accessible, many researches about factors affecting fuel consumptions have appeared.To sum up, the driving style of drivers, automobile emissions and the type of the routes are the mainly three factors.Classification is relatively single.In order to improve the classification accuracy’ this text studies that there exist some special roads,when drivers drive through them,their fuel consumptions will be similar because of road qualities.To achieve this goal, the first step is to calculate fuel consumptions per 100km of all city roads.Recognizing and examining special roads based on the ST-Matching algorithm.Third is to analyze road qualities of special roads like the length of roads, the speed of driving, and compare them with other common roads.Then we choose 4 cases to analyze. We find that when driving through special roads, fuel consumptions would be similar because of road qualities.Besides, the average length of special roads is longer than common roads’ and the mean velocity, the speed of getting and off special roads are faster than common roads’.The findings in this paper can filter out special road segments as noisy data in the study of relationship between driving styles and fuel consumptions,and it also has very high practical significance on recommending fuel-efficient paths.

The visual detection technology plays an important role in the industry of precision casting. With pertinence to the features of crack images of precision casting, an automatic detection and identification software for cracks of precision casting integrating crack segmentation, identification, measurements and extractions was designed and developed. Adopting the threshold segmentation, morphological processing, feature identifications as well as feature extraction and proposing a detection method with pertinence to the precision casting crack characteristics. The software effectively eliminates the influences on the detection results by the operators due to such factors as the detection environment, fatigue degree, visual resolution and working experience. The processing of the actually collected crack images of precision casting verified the efficacy and practicality of the software in the field of casting detection.

In this paper, a novel approach to change detection in synthetic aperture radar (SAR) images based on structure similarity (SSIM) and parametric kernel graph cuts is presented. Firstly, the SSIM is imported into change detection and a difference image constructed method based on SSIM is proposed. And then, the changed and unchanged pixels are identified from the difference image by the parametric kernel graph cuts algorithm. Experimental results obtained on real SAR images demonstrate the effectiveness of the proposed method.

The generation of 3D models for indoor objects and scenes is an attractive tool for digital city, virtual reality and SLAM purposes. Panoramic images are becoming increasingly more common in such applications due to their advantages to capture the complete environment in one single image with large field of view. The extraction and matching of image feature points are important and difficult steps in three-dimensional reconstruction, and ASIFT is a state-of-the-art algorithm to implement these functions. Compared with the SIFT algorithm, more feature points can be generated and the matching accuracy of ASIFT algorithm is higher, even for the panoramic images with obvious distortions. However, the algorithm is really time-consuming because of complex operations and performs not very well for some indoor scenes under poor light or without rich textures. To solve this problem, this paper proposes an improved ASIFT algorithm for indoor panoramic images: firstly, the panoramic images are projected into multiple normal perspective images. Secondly, the original ASIFT algorithm is simplified from the affine transformation of tilt and rotation with the images to the only tilt affine transformation. Finally, the results are re-projected to the panoramic image space. Experiments in different environments show that this method can not only ensure the precision of feature points extraction and matching, but also greatly reduce the computing time.

Many stereo matching algorithms use fixed color thresholds and a rigid cross skeleton to segment supports (viz., Cross method), which, however, does not work well for different images. To address this issue, this paper proposes a novel dual adaptive support (viz., DAS)-based stereo matching method, which uses both appearance and shape information of a local region to segment supports automatically, and, then, integrates the DAS-based cost aggregation with the absolute difference plus census transform cost, scanline optimization and disparity refinement to develop a stereo matching system. The performance of the DAS method is also evaluated in the Middlebury benchmark and by comparing with the Cross method. The results show that the average error for the DAS method 25.06% lower than that for the Cross method, indicating that the proposed method is more accurate, with fewer parameters and suitable for parallel computing.

Image matching between a blurred (caused by camera motion, out of focus, etc.) image and a non-blurred image is a critical task for many image/video applications. However, most of the existing local feature schemes fail to achieve this work. This paper presents a blur-invariant descriptor and a novel local feature scheme including the descriptor and the interest point detector based on moment symmetry – the authors’ previous work. The descriptor is based on a new concept - center peak moment-like element (CPME) which is robust to blur and boundary effect. Then by constructing CPMEs, the descriptor is also distinctive and suitable for image matching. Experimental results show our scheme outperforms state of the art methods for blurred image matching

Intelligent computer image processing technology provides convenience and possibility for designers to carry out designs. Shape analysis can be achieved by extracting SURF feature. However, high dimension of SURF feature causes to lower matching speed. To solve this problem, this paper proposed a fast fabric image matching algorithm based on SURF K-means and LSH algorithm. By constructing the bag of visual words on K-Means algorithm, and forming feature histogram of each image, the dimension of SURF feature is reduced at the first step. Then with the help of LSH algorithm, the features are encoded and the dimension is further reduced. In addition, the indexes of each image and each class of image are created, and the number of matching images is decreased by LSH hash bucket. Experiments on fabric image database show that this algorithm can speed up the matching and retrieval process, the result can satisfy the requirement of dress designers with accuracy and speed.

Gabor filters and K-means algorithm are two commonly used texture analysis methods. However, the texture feature vector has a high dimension by using Gabor filters, which will influence the operating efficiency. Meanwhile, K-means algorithm is affected by the initial clustering centers, and it may lead to the decrease of classification accuracy. Hence, Relief algorithm is applied to make a feature selection for Gabor texture feature, and obtain a suitable texture feature sunset. Furthermore, cuckoo search is used to optimize the clustering center of K-means algorithm, and enhance the accuracy and efficiency of texture recognition. Experimental results demonstrate the effectiveness of the proposed method.

An accurate measurement of large aviation part plays a key role in the assembly of aircraft. However, due to the limitation of spatial size, a calibration with large field of view and an accurate surface measurement of large part is hard to achieve. In this paper, an improved measurement method with spatial constraint calibration method and feature compression extraction method is proposed. Firstly, based on the proposed spatial constraint calibration method, the vision system is conveniently and precisely calibrated by using the designed SBA and SLT. Images of scanning laser stripes are captured by the calibrated cameras, simultaneously. Then the proposed feature compression extraction method is adopted to accurately extract centers of laser stripes. Finally, based on the binocular vision principle, the surface of part is reconstructed. The accuracy of proposed calibration method is verified in the lab. The results of the measurement of a standard part show the validity and precision of the proposed method.

The robust and distinctive local feature description is a critical component of image matching. In this paper, a novel descriptor, DGOH descriptor, is presented based on dual gradient orientation histogram. DGOH is a full rotation invariant descriptor, which takes advantage of spatial information of features by utilizing intensity order based subregion division method, meanwhile adopts the rotation invariant gradient calculation method in descriptor construction process. The discriminability of DGOH descriptor is enhanced by utilizing dominant and secondary gradient orientation histogram to represent the detected affine feature region. Performance evaluation experiments are carried out on the standard Oxford dataset. The experimental results show that the DGOH descriptor outperforms the state-of-the-art descriptors in terms of stability, precision and efficiency.

Feature extraction plays a key role in the classification performance of synthetic aperture radar automatic target recognition (SAR-ATR). It is very crucial to choose appropriate features to train a classifier, which is prerequisite. Inspired by the great success of Bag-of-Visual-Words (BoVW), we address the problem of feature extraction by proposing a novel feature extraction method for SAR target classification. First, Gabor based features are adopted to extract features from the training SAR images. Second, a discriminative codebook is generated using K-means clustering algorithm. Third, after feature encoding by computing the closest Euclidian distance, the targets are represented by new robust bag of features. Finally, for target classification, support vector machine (SVM) is used as a baseline classifier. Experiments on Moving and Stationary Target Acquisition and Recognition (MSTAR) public release dataset are conducted, and the classification accuracy and time complexity results demonstrate that the proposed method outperforms the state-of-the-art methods.

This paper proposes a visual perceptual descriptor (VPD) and a new approach to extract perceptual depth feature for 2D image retrieval. VPD mimics human visual system, which can easily distinguish regions that have different textures, whereas for regions which have similar textures, color features are needed for further differentiation. We apply VPD on the gradient direction map of an image, capture texture-similar regions to generate a VPD map. We then impose the VPD map on a quantized color map and extract color features only from the overlapped regions. To reflect the nature of perceptual distance in single 2D image, we propose and extract the perceptual depth feature by computing the nuclear norm of the sparse depth map of an image. Extracted color features and the perceptual depth feature are both incorporated to a feature vector, we utilize this vector to represent an image and measure similarity. We observe that the proposed VPD + depth method achieves a promising result, and extensive experiments prove that it outperforms other typical methods on 2D image retrieval.

To bridge the gap between the fuzziness of biometrics and the exactitude of cryptography, based on combining palmprint with two-layer error correction codes, a novel biometrics encryption method is proposed. Firstly, the randomly generated original keys are encoded by convolutional and cyclic two-layer coding. The first layer uses a convolution code to correct burst errors. The second layer uses cyclic code to correct random errors. Then, the palmprint features are extracted from the palmprint images. Next, they are fused together by XORing operation. The information is stored in a smart card. Finally, the original keys extraction process is the information in the smart card XOR the user’s palmprint features and then decoded with convolutional and cyclic two-layer code. The experimental results and security analysis show that it can recover the original keys completely. The proposed method is more secure than a single password factor, and has higher accuracy than a single biometric factor.

A novel optimal seam method which aims to stitch those images with overlapping area more seamlessly has been propos ed. Considering the traditional gradient domain optimal seam method and fusion algorithm result in bad color difference measurement and taking a long time respectively, the input images would be converted to HSV space and a new energy function is designed to seek optimal stitching path. To smooth the optimal stitching path, a simplified pixel correction and weighted average method are utilized individually. The proposed methods exhibit performance in eliminating the stitching seam compared with the traditional gradient optimal seam and high efficiency with multi-band blending algorithm.

Statistical signal processing is one of the most important tasks in a large amount of areas of scientific studies, such as astrophysics, geophysics, and space physics. Phase recurrence analysis and long-range persistence are the two dynamical structures of the underlying processes for the given natural phenomenon. Linear and nonlinear time series analysis approaches (cross-correlation analysis, cross-recurrence plot, wavelet coherent transform, and Hurst analysis) are combined to investigate the relative phase interconnection and long-range correlation between solar activity and geomagnetic activity for the time interval from 1932 January to 2017 January. The following prominent results are found: (1) geomagnetic activity lags behind sunspot numbers with a phase shift of 21 months, and they have a high level of asynchronous behavior; (2) their relative phase interconnections are in phase for the periodic scales during 8-16 years, but have a mixing behavior for the periodic belts below 8 years; (3) both sunspot numbers and geomagnetic activity can not be regarded as a stochastic phenomenon because their dynamical behaviors display a long-term correlation and a fractal nature. We believe that the presented conclusions could provide further information on understanding the dynamical coupling of solar dynamo process with geomagnetic activity variation, and the crucial role of solar and geomagnetic activity in the long-term climate change.

Shadow mapping is commonly used in real-time rendering. In this paper, we presented an accurate and efficient method of soft shadows generation from planar area lights. First this method generated a depth map from light’s view, and analyzed the depth-discontinuities areas as well as shadow boundaries. Then these areas were described as binary values in the texture map called binary light-visibility map, and a parallel convolution filtering algorithm based on GPU was enforced to smooth out the boundaries with a box filter. Experiments show that our algorithm is an effective shadow map based method that produces perceptually accurate soft shadows in real time with more details of shadow boundaries compared with the previous works.

The conventional Fourier ptychographic microscopy (FPM) is a computational imaging approach, which stitches together a sequence of low-resolution (LR) images captured by different angles illumination. However, the limitation of processing efficiency in capturing LR images is gradually becoming obvious. Utilizing the principle, aimed at reducing the amount of captured measurements and decreasing acquisition time, this paper proposes an optimized spectral sampling scheme. In this method, the importance of the spectra in the spectrum domain is analyzed and the more informative parts are selected. The acquisition efficiency can be increased because the selected images are captured and applied into the conventional FPM routine. Compared with the conventional FPM, experimental results significantly indicate that the redundancy of information and the time of image collection could decrease without debasing the quality of the reconstruction.

To better understand the mechanism of ejecting spray droplet and raindrop, a high-speed image analysis for droplet characterization technique was proposed in this study. In our design, three basic units containing a high-speed CCD camera, a light source, and automatic image acquisition and data processing unit, are integrated to assess the droplet motion. Next, a set of multistage image processing algorithms, the droplet detection and droplet tracking, were developed and conducted to analyze the droplet trajectories of the grabbed frames. By controlling a very short exposure time, the droplet size and velocity can be obtained to generate the droplet size distribution (DSD). To examine our proposed approaches, a commercially available spray nozzle and raindrop are used to evaluate the effectiveness. Meanwhile, the pair-matching rate is also involved to validate the recognition rate of the droplet tracking algorithm. Using this proposed framework, it can be used to provide valuable data to assess the spray droplet or the precipitation status by their droplet size to droplet velocity.

In this paper, a fast algorithm for image deblurring is discussed. The algorithm is based on generalized inverse iteration, and linearized Bregman iteration for the basis pursuit problem. Numerical experiments show that the chaotic algorithm for image restoration is effective and efficient.

With the development of barcodes for commercial use, people’s requirements for detecting barcodes by smart phone become increasingly pressing. The low quality of barcode image captured by mobile phone always affects the decoding and recognition rates. This paper focuses on locating and decoding EAN-13 barcodes in fuzzy images. We present a more accurate locating algorithm based on segment length and high fault-tolerant rate algorithm for decoding barcodes. Unlike existing approaches, location algorithm is based on the edge segment length of EAN -13 barcodes, while our decoding algorithm allows the appearance of fuzzy region in barcode image. Experimental results are performed on damaged, contaminated and scratched digital images, and provide a quite promising result for EAN -13 barcode location and decoding.

Automatic image annotation is now a tough task in computer vision, the main sense of this tech is to deal with managing the massive image on the Internet and assisting intelligent retrieval. This paper designs a new image annotation model based on visual bag of words, using the low level features like color and texture information as well as mid-level feature as SIFT, and mixture the pic2pic, label2pic and label2label correlation to measure the correlation degree of labels and images. We aim to prune the specific features for each single label and formalize the annotation task as a learning process base on Positive-Negative Instances Learning. Experiments are performed using the Corel5K Dataset, and provide a quite promising result when comparing with other existing methods.

At low bit rates, Block based transform coding method uses large quantization step to quantize transform coefficients, which usually causes compression artifacts for images. Post-processing strategy is a promising solution which can greatly improve the visual quality of degraded images without change of existing codec. In this paper, we propose an image deblocking method for JPEG compressed images using joint Gaussian mixture model (GMM) and anchored neighborhood regression priors. The proposed method takes advantage of image priors to reduce blocking artifacts and achieve a better image quality simultaneously. First, we utilize GMM to reduce blocking artifacts. Based on the assumption that similar image patches can be derived from one certain Gaussian probability distribution, we formulate the image deblocking as an optimization problem by maximizing a posteriori function. Solving this problem ultimately boils down to the liner Wiener filtering. We then learn mapping functions offline based on the recent adjusted anchored neighborhood regression to enhance image details and edges. Extensive experimental results validate that our proposed method performs better both objectively and subjectively compared to some recently presented methods.

Image restoration is a significant task in the fields of computer vision and image processing. Image restoration research consists of two aspects: kernel estimation and image restoration. A single image restoration method based on L1-regularized blur kernel estimation is proposed in this paper. First, a bilateral filter is used to remove the image noise effectively. Second, the improved shock filter is used to enhance the edge information of the image. Subsequently, L1-regularization method is used to estimate the blur kernel of the blurred image alternately, during which Split-Bregman algorithm is used to optimize the solution process. Finally, Hyper-Laplacian and sparse priors are applied to the image obtained from the non-blind deconvolution process. Experimental results show that compared to other methods, better restoration results as well as improved computational efficiency can be achieved with the proposed method.

Image annotation is a task of assigning semantic labels to an image. Recently, deep neural networks with visual attention have been utilized successfully in many computer vision tasks. In this paper, we show that conventional attention mechanism is easily misled by the salient class, i.e., the attended region always contains part of the image area describing the content of salient class at different attention iterations. To this end, we propose a novel attention shaping mechanism, which aims to maximize the non-overlapping area between consecutive attention processes by taking into account the history of previous attention vectors. Several weighting polices are studied to utilize the history information in different manners. In two benchmark datasets, i.e., PASCAL VOC2012 and MIRFlickr-25k, the average precision is improved by up to 10% in comparison with the state-of-the-art annotation methods.

Existing methods for visual saliency based image retrieval typically aim at single instance image. However, without any prior knowledge, the content of single instance image is ambiguous and these methods cannot effectively reflect the object of interest. In this paper, we propose a novel image retrieval framework based on multi-instance saliency model. First, the feature saliency is computed based on global contrast, local contrast and sparsity, and the synthesize saliency map is obtained by using Multi-instance Learning (MIL) algorithm to dynamically weight the feature saliency. Then we employ a fuzzy region-growth algorithm on the synthesize saliency map to extract the saliency object. We finally extract color and texture feature as the retrieval feature and measure feature similarity by Euclidean distance. In the experiments, the proposed method can achieve higher multi-instance image retrieval accuracy than the other single instance image retrieval methods based on saliency model.

A novel single image dedusting method based on non-overlap stitching is investigated to solve the issue of image degradation in the non-uniform dusty environment with multiple scattering light. First, the first-order multiple scattering model is used to produce recovered images of the original dusty image by different transmission values. Then each recovered image is classified to clear and dusty regions using an improved dust detection operation by which the clear regions are kept. By repeating the first two steps and by reducing transmission value from 1 to the terminal value optimized by a cost function, series clear regions are produced. Finally, all clear regions with the rest dusty regions in the last recovered image are stitched into a restored image. Experimental results show that this method can effectively remove dust in the image and significantly improve the image quality obviously.

In this paper, we study how to compactly represent image’s characteristics for fast image retrieval. We propose supervised rotation invariant compact discriminative binary descriptors through combining convolutional neural network with hashing. In the proposed network, binary codes are learned by employing a hidden layer for representing latent concepts that dominate on class labels. A loss function is proposed to minimize the difference between binary descriptors that describe reference image and the rotated one. Compared with some other supervised methods, the proposed network doesn’t have to require pair-wised inputs for binary code learning. Experimental results show that our method is effective and achieves state-of-the-art results on the CIFAR-10 and MNIST datasets.

Combined 3D laser scanner and monocular camera is an important way to reconstruct 3D environment. This paper presents a new fusion scheme which takes more comprehensive use of heterogeneous data from different sensors. First, we extract the 3D structure information from sequence images to aid initial registration which provides a sufficiently accurate pose estimation for an ICP algorithm to perform the fine alignment. Second, extracting points from sequence images by dense reconstruction, registering the heterogeneous data that can supplement the details of model and solve the problem of ambiguity of images. The efficiency of the presented method has been tested on simulation software we programmed which simulates the process of heterogeneous data acquired and model reconstruction. The results show that the initial registration can acquired an accuracy and stable alignment for ICP and reconstruct more accuracy model, and this method does not need joint calibration of laser scanner and camera or manual intervention, and has better adaptability.

To improve the results of image fusion, taking multi-focus color image as the research object, a self-adaptive pulse coupled neural network multi-focus color image fusion method is presented, which is based on the nonsubsampled contourlet transform (NSCT). And compute the hue component and saturation component of the fused image by the fused intensity component, and finish the whole fused process of the multi-focus color image. The results show that the proposed image fusion method has a better quality than the fused by the wavelet transform and the nonsubsampled contourlet transform, as the same time the fused results in HSI model is of better quality than that fused in RGB model.

This paper proposed a weighted l₀ shearlet-based model for image deblurring. The main purpose of this work is to further exploiting the sparsity of the reconstructed signal. In order to achieve this goal, a generalized gradient regularizer is introduced to the model. The added regularizer can suppress artifacts effectively. The split Bregman algorithm is used to update the multi-scale weighted matrix in the each iteration. This weighted matrix can transmit the solution information in the present step to the next step by support detection. According to this procedure, the whole algorithm framework forms a learning process. Experimental results suggest that the proposed algorithm yields significantly improvement in terms of PSNR. However, it also shows that more computing time is required due to the utilization of the redundant shearlet system.

New grayscale morphological operators on hypergraph are proposed to avoid the loss of details caused by fixed structure element effectively. Hypergraph, the most general structure in discrete mathematics, is also a subset of a finite set. Being a structured representation of information, the ordinary image can be transformed into a hypergraph model, which can integrate hypergraph theory with mathematical morphology theory. Because hypergraphs have good performance in structuring information, first of all, this paper designs a reasonable method of turning grayscale images into hypergraph space. Then based on hypergraph theory, new grayscale morphological operators on hypergraph are defined. Experiments show that using the new operators can avoid the loss of image detail information, and improve the precision of image processing.

Tank, as a vital ground weapon, plays an irreplaceable role in the war. The article did the research of infrared image of the tank. Firstly, the 3D model of tank was established. And then the infrared radiation model of the target was constructed by analysing the infrared characteristics of the tank’s different parts.. Finally the infrared radiation value of the tank under different states was calculated and the simulation of infrared characteristics of the tank under different states was done, which will provide reference for the research on infrared characteristics of the army's battlefield target.

In this paper, we propose an adaptive mean filter FCM algorithm based on LAB transform to improve the anti-noise ability of the traditional FCM in image segmentation, which lack of using spatial information. The algorithm firstly counts the peak value of the whole image pixel, and then weights the transformed color component according to the statistical result, and gives the different color space different information amount. At the same time, the algorithm combines the variance of the image to filter the image, and adjusts the filter degree adaptively to use the spatial information most effectively. The experimental results show that our algorithm is better than the traditional algorithm in segmentation effect, and has great improvement in anti-noise ability.

The human vision system has abilities for feature detection, learning and selective attention with some properties of hierarchy and bidirectional connection in the form of neural population. In this paper, a multiscale Markov random field model in the wavelet domain is proposed by mimicking some image processing functions of vision system. For an input scene, our model provides its sparse representations using wavelet transforms and extracts its topological organization using MRF. In addition, the hierarchy property of vision system is simulated using a pyramid framework in our model. There are two information flows in our model, i.e., a bottom-up procedure to extract input features and a top-down procedure to provide feedback controls. The two procedures are controlled simply by two pyramidal parameters, and some Gestalt laws are also integrated implicitly. Equipped with such biological inspired properties, our model can be used to accomplish different image segmentation tasks, such as edge detection and region segmentation.

Because of the only use of local spatial features, classical methods for transition region extraction and thresholding would result in unsatisfied, even complete failure results under the existence of noises or outliers. In view of this, we propose a novel algorithm based on nonlocal spatial feature and gray level difference. This algorithm generates the nonlocal spatial feature and gray level difference first, and constructs the effective feature matrix based on the above two features, then obtains an automatic threshold related to the effective feature matrix according to a statistical method for thresholding, meanwhile extracts the transition region. Finally, the algorithm obtains the optimal grayscale threshold by calculating the grayscale mean of transition pixels, and yields the binary result. Experimental results show that, the proposed algorithm performs good result of transition region extraction and thresholding, and it is reasonable and effective, can be as an alternative to traditional methods.

Overlapping cell segmentation is a prerequisite for the analysis of cervical smear images. Because of the complexity of overlapping situations and the poor contrast of overlapping edges, this problem is one of the most challenges in this field. In this paper, a novel unsupervised segmentation method without needing the training data for overlapping cervical smear images is proposed. First, this method uses a kind of graph cuts to separate all cell clumps from the background. A cell clump may contain the different number of cervical cells. Second, each clump is segmented into non-overlapping regions as rough cells using Voronoi diagram. Third, in order to refine the segmentation of overlapping regions, a minimum enclosing ellipse is used to fit in each rough cell and the overlapped parts of each cell are replaced with the relational regions in this ellipse. Finally, the above overlapped parts and the connected parts of the Voronoi rough cell are merged to form a complete cell. Experiments are conducted on 2 publicly released ISBI datasets and results show that the proposed segmentation method achieves the state-of-art performance.

Optical coherence tomography (OCT) is a high resolution and non-invasive imaging modality that has become one of the most prevalent techniques for ophthalmic diagnostic. However, manual segmentation is often a time-consuming and subjective process. In this work, we present a new method for retinal layer segmentation in retinal optical coherence tomography images, which uses a deep convolutional feature to train a structured random forest classifier. The experimental results show that our method achieves good results with the mean distance error of 1.45 pixels whereas that of the state-of-the-art was 1.68 pixels, and achieve a F-score of 0.86 which is also better than 0.83 that is obtained by the state-of-the-art method.

In the retinal image, characteristics of fluid have great significance for diagnosis in eye disease. In the clinical, the segmentation of fluid is usually conducted manually, but is time-consuming and the accuracy is highly depend on the expert’s experience. In this paper, we proposed a segmentation method based on convolution neural network (CNN) for segmenting the fluid from fundus image. The B-scans of OCT are segmented into layers, and patches from specific region with annotation are used for training. After the data set being divided into training set and test set, network training is performed and a good segmentation result is obtained, which has a significant advantage over traditional methods such as threshold method.

The identification of white blood cells was important as it provided diagnosis information of different kinds of diseases. However, traditional light microscopy based leukocyte cells recognition and segmentation methods usually inaccurate. This paper proposed a hybrid algorithm applied mathematical support vector machine cells screening algorithm combined with BandMax and spectral angle mapping for white blood cell segmentation, that was, it treated BandMax and spectral angle mapping as a new preprocessing method to divide the boundaries between cells, and then used support vector machine cells screening algorithm to segment the hyperspectral cell images more efficiently and precisely than traditional segmentation algorithms. Experimental results shown that the hybrid algorithm provided higher classification accuracy than traditional methods on improving the classification accuracy and effective extraction of white blood cells. By combing both spatial and spectral features, this strategy had been successfully tested for classifying objects among leukocytes, erythrocytes and serums in raw samples, including spectral features reached higher accuracy than any single algorithm cases, with a maximum improvement of nearly 26.82%.

As an imaging technology used in remote sensing, hyperspectral imaging can provide more information than traditional optical imaging of blood cells. In this paper, an AOTF based microscopic hyperspectral imaging system is used to capture hyperspectral images of blood cells. In order to achieve the segmentation of red blood cells, Gaussian process using squared exponential kernel function is applied first after the data preprocessing to make the preliminary segmentation. The derivative spectrum with spectral angle mapping algorithm is then applied to the original image to segment the boundary of cells, and using the boundary to cut out cells obtained from the Gaussian process to separated adjacent cells. Then the morphological processing method including closing, erosion and dilation is applied so as to keep adjacent cells apart, and by applying median filtering to remove noise points and filling holes inside the cell, the final segmentation result can be obtained. The experimental results show that this method appears better segmentation effect on human red blood cells.

Cancer recognition is the prerequisite to determine appropriate treatment. This paper focuses on the semantic segmentation task of microvascular morphological types on narrowband images to aid clinical examination of esophageal cancer. The most challenge for semantic segmentation is incomplete-labeling. Our key insight is to build fully convolutional networks (FCNs) with double-label to make pixel-wise predictions. The roi-label indicating ROIs (region of interest) is introduced as extra constraint to guild feature learning. Trained end-to-end, the FCN model with two target jointly optimizes both segmentation of sem-label (semantic label) and segmentation of roi-label within the framework of self-transfer learning based on multi-task learning theory. The learning representation ability of shared convolutional networks for sem-label is improved with support of roi-label via achieving a better understanding of information outside the ROIs. Our best FCN model gives satisfactory segmentation result with mean IU up to 77.8% (pixel accuracy > 90%). The results show that the proposed approach is able to assist clinical diagnosis to a certain extent.

Since the image acquisition is interfered by various factors, and the image of the license plate is often distorted, the distortion correction of the license plate image is of great significance to improve the accuracy of the license plate recognition. In this paper, we propose a new method to segment distorted images for linear approximation. The main idea is to segment the non-linear distorted license plate images into linear parts according to the license plate character, searching for the corners of each part, then reconstructing the non-linear distortion of the license plate image by using the perspective projection model, finally realizing the license plate image correction. Experimental results show that the proposed method is effective for the correction of non-linear distortion plate images.

In this paper, we proposed a novel method for image segmentation by adding dynamic weighting coefficients based on the original fuzzy weighting in combination with two-dimensional gray-level histogram segmentation method in fuzzy cmeans algorithm (FCM). Finally, post-segmentation is carried out to improve the quality of image segmentation. Experimental results show that this method can speed up the process of finding the optimal solution and improve the image segmentation accuracy and robustness of anti-noise.

While the popular thin layer scanning technology of spiral CT has helped to improve diagnoses of lung diseases, the large volumes of scanning images produced by the technology also dramatically increase the load of physicians in lesion detection. Computer-aided diagnosis techniques like lesions segmentation in thin CT sequences have been developed to address this issue, but it remains a challenge to achieve high segmentation efficiency and accuracy without much involvement of human manual intervention. In this paper, we present our research on automated segmentation of lung parenchyma with an improved geodesic active contour model that is geodesic active contour model based on similarity (GACBS). Combining spectral clustering algorithm based on Nystrom (SCN) with GACBS, this algorithm first extracts key image slices, then uses these slices to generate an initial contour of pulmonary parenchyma of un-segmented slices with an interpolation algorithm, and finally segments lung parenchyma of un-segmented slices. Experimental results show that the segmentation results generated by our method are close to what manual segmentation can produce, with an average volume overlap ratio of 91.48%.

According to features of micro-seismic signal, this paper defines the similarity between signals using the cross-correlation, and applies bilateral filtering to the domain of micro- seismic profile image processing. Relative to the general bilateral filtering and curvelet filtering, experimental results confirm that the bilateral filtering based on cross-correlation can remove interference noises a lot, enhance the signal-to-noise ration, and keep the continuity of horizontal micro-seismic events, that will ensure the accuracy of micro-seismic event identification and positioning.

To overcome the problem of image de-noising, an efficient image de-noising approach based on mathematical morphology and multi-objective particle swarm optimization (MOPSO) is proposed in this paper. Firstly, constructing a series and parallel compound morphology filter based on open-close (OC) operation and selecting a structural element with different sizes try best to eliminate all noise in a series link. Then, combining multi-objective particle swarm optimization (MOPSO) to solve the parameters setting of multiple structural element. Simulation result shows that our algorithm can achieve a superior performance compared with some traditional de-noising algorithm.

This paper introduces a fractional order total variation (FOTV) based model with three different weights in the fractional order derivative definition for multiplicative noise removal purpose. The fractional-order Euler Lagrange equation which is a highly non-linear partial differential equation (PDE) is obtained by the minimization of the energy functional for image restoration. Two numerical schemes namely an iterative scheme based on the dual theory and majorization-- minimization algorithm (MMA) are used. To improve the restoration results, we opt for an adaptive parameter selection procedure for the proposed model by applying the trial and error method. We report numerical simulations which show the validity and state of the art performance of the fractional-order model in visual improvement as well as an increase in the peak signal to noise ratio comparing to corresponding methods. Numerical experiments also demonstrate that MMAbased methodology is slightly better than that of an iterative scheme.

Microscopic images provide lots of useful information for modern diagnosis and biological research. However, due to the unstable lighting condition during image capturing, two main problems, i.e., high-level noises and low image contrast, occurred in the generated cell images. In this paper, a simple but efficient enhancement framework is proposed to address the problems. The framework removes image noises using a hybrid method based on wavelet transform and fuzzy-entropy, and enhances the image contrast with an adaptive morphological approach. Experiments on real cell dataset were made to assess the performance of proposed framework. The experimental results demonstrate that our proposed enhancement framework increases the cell tracking accuracy to an average of 74.49%, which outperforms the benchmark algorithm, i.e., 46.18%.

Recently, T. Celik proposed an effective image contrast enhancement (CE) method based on spatial mutual information and PageRank (SMIRANK). According to the state-of-the-art evaluation criteria, it achieves the best visual enhancement quality among existing global CE methods. However, SMIRANK runs much slower than the other counterparts, such as histogram equalization (HE) and adaptive gamma correction. Low computational complexity is also required for good CE algorithms. In this paper, we novelly propose a fast SMIRANK algorithm, called FastSMIRANK. It integrates both spatial and gray-level downsampling into the generation of pixel value mapping function. Moreover, the computation of rank vectors is speeded up by replacing PageRank with a simple yet efficient row-based operation of mutual information matrix. Extensive experimental results show that the proposed FastSMIRANK could accelerate the processing speed of SMIRANK by about 20 times, and is even faster than HE. Comparable enhancement quality is preserved simultaneously.

Retinal vessel detection is an essential part of the computer-aided diagnosis of eye diseases. Due to non-perfect imaging environment, retinal images often appear with intensity variations and artificial noises. This work proposes a two-step nonlinear retinal image enhancement to compensate for those imperfections of retinal images. The first step reduces intensity fluctuations of the image and the second step attenuates impulsive noise while preserving retinal vessels. Classification on the feature vector extracted from the enhanced retinal images is performed by using a linear SVM classifier. Experimental results demonstrate that the proposed method of two-step nonlinear image enhancement visibly improves the vessel detection performance, achieving better accuracy than that without enhancement process on the both DRIVE and STARE databases.

Based on good amplitude frequency characteristics and the spatial global correlation of fractional-order differential, an energy functional of haze image enhancement is established by taking fractional derivative on both sides of the atmospheric physics scattering model, and a haze image enhancement model based on space fractional-order partial differential equation is obtained by using steepest descent method. Based on fast wavelet transform, the low-frequency part of patch transmission and the high-frequency part of point transmission are fused to estimate the transmission. Finally, the numerical solution of the fractional-order partial differential equation is obtained by the finite difference method. The experimental results show that the algorithm can improve the contrast, brightness and clarity of the image, and it is an effective image enhancement method for haze images.

In view interpolation, information missing often exists in initial depth map, moreover, disocclusion regions usually occur along the foreground object boundaries after 3D warping. Generally, initial depth map and warped depth map have a strong influence on the performance of view interpolation. However, most of existing view interpolation algorithms only emphasize hole filling of the warped color image. In this paper, a superpixel-based method is proposed for initial depth map enhancement and warped depth map hole filling. Firstly, the color image is segmented using simple linear iterative clustering (SLIC) algorithm, and after that, the associated depth map is segmented with the same label. Then, the depthmissing pixels are recovered by considering color and depth superpixel information jointly. Additionally, holes of the disocclusion regions in the warped depth map can also be filled efficiently via superpixel-based segmentation. Experimental results show that with the proposed method the quality of the interpolated view has been improved significantly in terms of both subjective and objective evaluations.

This paper presents a preliminary study of the effect of empty bins on image upscaling in capsule endoscopy. The presented study was conducted based on results of existing contrast enhancement and interpolation methods. A low contrast enhancement method based on pixels consecutiveness and modified bilinear weighting scheme has been developed to distinguish between necessary empty bins and unnecessary empty bins in the effort to minimize the number of empty bins in the input image, before further processing. Linear interpolation methods have been used for upscaling input images with stretched histograms. Upscaling error differences and similarity indices between pairs of interpolation methods have been quantified using the mean squared error and feature similarity index techniques. Simulation results demonstrated more promising effects using the developed method than other contrast enhancement methods mentioned.

The enhancement of Low Light Level Images (LLLIs) is challenging, due to their poor brightness and low contrast. Traditional enhancement methods fail to perform satisfactorily when applying to LLLIs. In this paper, we formulate the LLLI enhancement as a regression problem: the regressor maps patches of input image to enhanced patches, and the regression function is estimated by learning from sample images. We implemented two efficient regression methods based on piecewise linear regression: locally linear regression and random forest (RF). Meanwhile, we designed a new split function considering reconstruction error for random forest method. Experimental results on an open dataset and practical LLLIs demonstrate the effectiveness of our methods. The RF regression method performs superiorly in both enhancement quality and computation efficiency

This paper proposes an image detection model to detect and classify supermarkets shelves’ commodity. Based on the principle of the features directly affects the accuracy of the final classification, feature maps are performed to combine high level features with bottom level features. Then set some fixed anchors on those feature maps, finally the label and the position of commodity is generated by doing a box regression and classification. In this work, we proposed a model named Deconvolutiuon Single Shot MultiBox Detector, we evaluated the model using 300 images photographed from real supermarket shelves. Followed the same protocol in other recent methods, the results showed that our model outperformed other baseline methods.

There have been noticed growing explorers' interest in drawing conclusions based on information of data coded in a graphic form. The neuronal identification of pictorial data, with special emphasis on both quantitative and qualitative analysis, is more frequently utilized to gain and deepen the empirical data knowledge. Extraction and then classification of selected picture features, such as color or surface structure, enables one to create computer tools in order to identify these objects presented as, for example, digital pictures. The work presents original computer system “Processing the image v.1.0” designed to digitalize pictures on the basis of color criterion. The system has been applied to generate a reference learning file for generating the Artificial Neural Network (ANN) to identify selected kinds of butterflies from the Papilionidae family.

Considering that the manual inspection of the yarn-dyed fabric can be time consuming and less efficient, a convolutional neural network (CNN) solution based on the modified AlexNet structure for the classification of the yarn-dyed fabric defect is proposed. CNN has powerful ability of feature extraction and feature fusion which can simulate the learning mechanism of the human brain. In order to enhance computational efficiency and detection accuracy, the local response normalization (LRN) layers in AlexNet are replaced by the batch normalization (BN) layers. In the process of the network training, through several convolution operations, the characteristics of the image are extracted step by step, and the essential features of the image can be obtained from the edge features. And the max pooling layers, the dropout layers, the fully connected layers are also employed in the classification model to reduce the computation cost and acquire more precise features of fabric defect. Finally, the results of the defect classification are predicted by the softmax function. The experimental results show the capability of defect classification via the modified Alexnet model and indicate its robustness.

To improve the feature recognition ability of deep model transfer learning, we propose a hybrid deep transfer learning method for image classification based on restricted Boltzmann machines (RBM) and convolutional neural networks (CNNs). It integrates learning abilities of two models, which conducts subject classification by exacting structural higher-order statistics features of images. While the method transfers the trained convolutional neural networks to the target datasets, fully-connected layers can be replaced by restricted Boltzmann machine layers; then the restricted Boltzmann machine layers and Softmax classifier are retrained, and BP neural network can be used to fine-tuned the hybrid model. The restricted Boltzmann machine layers has not only fully integrated the whole feature maps, but also learns the statistical features of target datasets in the view of the biggest logarithmic likelihood, thus removing the effects caused by the content differences between datasets. The experimental results show that the proposed method has improved the accuracy of image classification, outperforming other methods on Pascal VOC2007 and Caltech101 datasets.

A CAPTCHA (“Completely Automated Public Turing test to tell Computers and Human Apart”) system is a program that most humans can pass but current computer programs could hardly pass. As the most common type of CAPTCHAs , text-based CAPTCHA has been widely used in different websites to defense network bots. In order to breaking textbased CAPTCHA, in this paper, two trained CNN models are connected for the segmentation and classification of CAPTCHA images. Then base on these two models, we apply sliding window segmentation and voting classification methods realize an end-to-end CAPTCHA breaking system with high success rate. The experiment results show that our method is robust and effective in breaking text-based CAPTCHA with noise.

In order to cater to the evidentiary requirements of the text image, this paper proposes a fragile watermarking algorithm based on Hash function and Henon map. The algorithm is to divide a text image into parts, get flippable pixels and nonflippable pixels of every lump according to PSD, generate watermark of non-flippable pixels with MD5-Hash, encrypt watermark with Henon map and select embedded blocks. The simulation results show that the algorithm with a good ability in tampering localization can be used to authenticate and forensics the authenticity and integrity of text images

This paper proposes a color separation algorithm for glass printing. The algorithm separates the original image into six kinds of color layers and can be easily applied to n kinds of arbitrary colors’ case. Given six kinds of ink, this algorithm mixes some of them up averagely to reach a variety of basic colors. The transparency channel is considered in the separation process, and any part of the image can be adjusted separately by setting its alpha value. After dealing with the transparency channel, color is separated by dithering method and the corresponding color layers are output at the same time. Moreover, the whole color separation process does not need any pre-treatment or post-treatment to do. Experiment shows that the original image can be appropriately restored on glass by using this algorithm and the transparency of the original image is well-expressed.

The signal recorded by a low resolution hyperspectral remote sensor from a given pixel, letting alone the effects of the complex terrain, is a mixture of substances. To improve the accuracy of classification and sub-pixel object detection, hyperspectral unmixing(HU) is a frontier-line in remote sensing area. Unmixing algorithm based on geometric has become popular since the hyperspectral image possesses abundant spectral information and the mixed model is easy to understand. However, most of the algorithms are based on pure pixel assumption, and since the non-linear mixed model is complex, it is hard to obtain the optimal endmembers especially under a highly mixed spectral data. To provide a simple but accurate method, we propose a minimum volume simplex analysis constrained (MVSAC) unmixing algorithm. The proposed approach combines the algebraic constraints that are inherent to the convex minimum volume with abundance soft constraint. While considering abundance fraction, we can obtain the pure endmember set and abundance fraction correspondingly, and the final unmixing result is closer to reality and has better accuracy. We illustrate the performance of the proposed algorithm in unmixing simulated data and real hyperspectral data, and the result indicates that the proposed method can obtain the distinct signatures correctly without redundant endmember and yields much better performance than the pure pixel based algorithm.

The traditional tone mapping algorithm for the display of high dynamic range (HDR) image has the drawback of losing the impression of brightness, contrast and color information. To overcome this phenomenon, we propose a new tone mapping algorithm based on dividing the image into different exposure regions in this paper. Firstly, the over-exposure region is determined using the Local Binary Pattern information of HDR image. Then, based on the peak and average gray of the histogram, the under-exposure and normal-exposure region of HDR image are selected separately. Finally, the different exposure regions are mapped by differentiated tone mapping methods to get the final result. The experiment results show that the proposed algorithm achieve the better performance both in visual quality and objective contrast criterion than other algorithms.

The precision of training samples has important influence on the accuracy of Supervised Classification of Remote Sensing Images. It is easy to omit the ground feature which covers a small amount by visual selection directly on the original image. This leads to a large number of wrong points and leakage points which influences the Classification Accuracy of Remote Sensing Images. Based on what was said above, we first use K-means clustering algorithm to initialize the samples in this paper. According to the clustering results, the training samples selected by visual labeling and density-based k-means algorithm participated in the SVM classification model. The results show that selecting training samples by the k-means algorithm based on density can reduce the influence of visual labeling on the subjective factors of training sample selection. Meanwhile,through the optimization of the radius weighting factor, we have found optimal combination of weighting factors and then the classification accuracy of remote sensing images can be increased from 81.70% to 88.89%.

In the recent years, there has been a continuously increasing demand for vegetables and dried vegetables. This trend affects the growth of the dehydration industry in Poland helping to exploit excess production. More and more often dried vegetables are used in various sectors of the food industry, both due to their high nutritional qualities and changes in consumers’ food preferences. As we observe an increase in consumer awareness regarding a healthy lifestyle and a boom in health food, there is also an increase in the consumption of such food, which means that the production and crop area can increase further. Among the dried vegetables, dried carrots play a strategic role due to their wide application range and high nutritional value. They contain high concentrations of carotene and sugar which is present in the form of crystals. Carrots are also the vegetables which are most often subjected to a wide range of dehydration processes; this makes it difficult to perform a reliable qualitative assessment and classification of this dried product. The many qualitative properties of dried carrots determining their positive or negative quality assessment include colour and shape. The aim of the research was to develop and implement the model of a computer system for the recognition and classification of freeze-dried, convection-dried and microwave vacuum dried products using the methods of computer image analysis and artificial neural networks.

The Bridges had collapsed accidents in recent years due to bridges quality problems. Therefore, concretes nondestructive testing are particularly important. At present, most applications are Ground Penetrating Radar (GPR) technology in the detection of reinforced concretes structure. GPR are used the pulse method which alongside with definitive advantages, but the testing of the internal structure of the small thickness concretes has very low resolution by this method. In this paper, it’s the first time to use the ultra-wideband (UWB) stepped frequency conversion radar above problems. We use vector network analyzer and double ridged horn antenna microwave imaging system to test the reinforced concretes block. The internal structure of the concretes is reconstructed with a method of synthetic aperture of ω-k algorithm. By this method, the depth of the steel bar with the diameter of 1cm is shown exactly in the depth of 450mm×400mm×500mm and the depth error do not exceed 1cm.

This paper proposes a novel imaging technique which combines clustering sub-dictionary learning and gradient histogram preservation to improve the quality of compressive imaging from two aspects: edge sharpness and noise suppression. Practical experiments further demonstrate better results on practical optical imaging application in terms of weighted peak signal-to-noise ratio and measure of feature similarity index.

A pre-processing step is needed to correct for the bias field signal before submitting corrupted MR images to such image-processing algorithms. This study presents a new bias field correction method. The method creates a Gaussian multi-scale space by the convolution of the inhomogeneous MR image with a two-dimensional Gaussian function. In the multi-Gaussian space, the method retrieves the image details from the differentiation of the original image and convolution image. Then, it obtains an image whose inhomogeneity is eliminated by the weighted sum of image details in each layer in the space. Next, the bias field-corrected MR image is retrieved after the Υ correction, which enhances the contrast and brightness of the inhomogeneity-eliminated MR image. We have tested the approach on T1 MRI and T2 MRI with varying bias field levels and have achieved satisfactory results. Comparison experiments with popular software have demonstrated superior performance of the proposed method in terms of quantitative indices, especially an improvement in subsequent image segmentation.

Radar correlated imaging (RCI) introduces the optical correlated imaging technology to traditional microwave imaging, which has raised widespread concern recently. Conventional RCI methods neglect the structural information of complex extended target, which makes the quality of recovery result not really perfect, thus a novel combination of negative exponential restraint and total variation (NER-TV) algorithm for extended target imaging is proposed in this paper. The sparsity is measured by a sequential order one negative exponential function, then the 2D total variation technique is introduced to design a novel optimization problem for extended target imaging. And the proven alternating direction method of multipliers is applied to solve the new problem. Experimental results show that the proposed algorithm could realize high resolution imaging efficiently for extended target.

In this paper, a technology of multi-radar imaging based on compressed sensing is proposed to improve image resolution. By constructing sample matrix, multi-radar super-resolution imaging is transformed to a compressed sensing problem. Utilizing the signal’s sparsity, super-resolution image can be obtained by solving an optimization problem. Simulation shows effectiveness of this technology.

Raw scanned 3D point clouds are usually irregularly distributed due to the essential shortcomings of laser sensors, which therefore poses a great challenge for high-quality 3D surface reconstruction. This paper tackles this problem by proposing a local hierarchical clustering (LHC) method to improve the consistency of point distribution. Specifically, LHC consists of two steps: 1) adaptive octree-based decomposition of 3D space, and 2) hierarchical clustering. The former aims at reducing the computational complexity and the latter transforms the non-uniform point set into uniform one. Experimental results on real-world scanned point clouds validate the effectiveness of our method from both qualitative and quantitative aspects.

In this paper, we proposed a new method to improve the resolution of depth image. This method uses a high-resolution color image as a reference. We also use the bilateral filtered image as the guide image, and obtains the reconstructed image through joint bilateral filter and guidance filter. The experimental results show that this method can enhance the spatial resolution of the depth image and effectively eliminate the ringing effect and artifacts.

In recent years, deep convolutional neural networks come into use in image inpainting and super-resolution in many fields. Distinct to most of the former methods requiring to know beforehand the local information for corrupted pixels, we propose a 20-depth fully convolutional network to learn an end-to-end mapping a dataset of damaged/ground truth subimage pairs realizing non-local blind inpainting and super-resolution. As there often exist image with huge corruptions or inpainting on a low-resolution image that the existing approaches unable to perform well, we also share parameters in local area of layers to achieve spatial recursion and enlarge the receptive field. To avoid the difficulty of training this deep neural network, skip-connections between symmetric convolutional layers are designed. Experimental results shows that the proposed method outperforms state-of-the-art methods for diverse corrupting and low-resolution conditions, it works excellently when realizing super-resolution and image inpainting simultaneously

According to the sequence of license plate image has low resolution, in order to improve the image resolution, this paper proposes a multiframe super-resolution image reconstruction algorithm based on overcome registration error and Gauss weight interpolation. First, multiframe image sub-pixel image registration, through extracting the SIFT keypoints of the image, rough matching with BBF algorithm, accurate matching with RANSAC algorithm, establishing the homography matrix between matching points. In order to overcome the registration error, this paper proposes a method based on threshold to overcome the registration error. Finally, the multiframe low resolution image is projected into the high resolution image grid by image registration. The value of the high resolution grid is reconstructed with a non-uniform interpolation algorithm based on Gauss weight. The experiments show that both on the subjective evaluation and objective evaluation can obtain satisfactory reconstruction results, and the registration error is robust.

The rapid development of the satellite observation technology provides a very rich source of data for sea reconnaissance and ships surveillance. In the face of such a vast sea remote sensing data, it is urgent need to realize the automatic ship detection in optical remote sensing images, but the optical remote sensing images are easily affected by meteorological conditions, such as clouds, waves, which results in larger false alarm; and the weak contrast between optical remote sensing image target and background is easy to cause missing alarm. In this paper, a novel algorithm based on wavelet transform for ship target detection in optical remote sensing images is proposed, which can effectively remove these noise and interference. The segmentation of sea and land background is first applied to the image preprocessing to achieve more accurate detection results, and then discrete wavelet transform is used to deal with the part of sea background. The results show that almost all of the offshore ships can be detected, and through the comparison of the results of four different wavelet basis functions, the accuracy of ship detection is further improved.

Understanding a scene provided by Very High Resolution (VHR) satellite imagery has become a more and more challenging problem. In this paper, we propose a new method for scene classification based on different pre-trained Deep Features Learning Models (DFLMs). DFLMs are applied simultaneously to extract deep features from the VHR image scene, and then different basic operators are applied for features combination extracted with different pre-trained Convolutional Neural Networks (CNN) models. We conduct experiments on the public UC Merced benchmark dataset, which contains 21 different areal categories with sub-meter resolution. Experimental results demonstrate the effectiveness of the proposed method, as compared to several state-of-the-art methods.

In military fields, it is essential to monitor the dynamic of aircrafts through remote sensing images. Due to lack of automated assistive analysis methods of recent works, we propose a novel method for automatically monitoring the dynamic of aircrafts through remote sensing images in this paper. The method consists of two phases: (i) establish a priori model of aircrafts in airports and learn a Convolutional Neural Networks (CNN) classifier that identifies the state of aircrafts, and (ii) predict the states of aircrafts in the new images. The proposed method was tested on the remote sensing images of two typical airports. Experimental results show that the method is able to monitor the dynamic of aircrafts with high accuracy. We conclude that the method can report the states of aircrafts in airports correctly.

For remote sensing images with rich content of features, this paper presents a remote sensing image classification algorithm based on twin support vector machine (TWSVM) and multi-feature optimization. Firstly, we extract color feature and shape feature of remote sensing image and introduce the local angular phase (LAP) histogram, which has high texture description ability, as the texture feature. Due to these three kinds of feature represent different emphases of remote sensing image; the reasonable combination of them can be more comprehensive description of the contents of remote sensing image. Secondly, we use kernel principal component analysis (KPCA) to reduce the dimension of every kind of feature, and construct reasonable feature space based on different weights obtained by the distribution of feature space. Finally, the remote sensing image samples classification and test is completed in the TWSVM model that has better classification performance. Experimental results on the USGS test-set show that, the average classification accuracy of the proposed algorithm is reached to 93.7% compared with three popular methods. Compared with the highest classification accuracy of single feature classification algorithm, the average classification accuracy of the proposed algorithm has been improved by 16%, 14.5% and 9.2%.

Land surface temperature is playing a very important role in ground-atmosphere interaction. It is a key parameter in the global change studies. So it becomes one of important tasks in quantitative remote sensing research to make use of satellite remote sensing to retrieve land surface temperature at present. Land surface temperature is a crucial parameter of thermal energy distribution in region, the most efficient way to acquire it is through thermal infrared remote sensing images. On the basis of studies on the land surface temperature. With the support of remote sensing and GIS technology, using the single window algorithm to quantitative inversion of thermal infrared remote sensing image, obtain the spatial distribution of surface temperature change in the study area

Remote sensing system fitted on Unmanned Aerial Vehicle (UAV) can obtain clear images and high-resolution aerial photographs. It has advantages of strong real-time, flexibility and convenience, free from influence of external environment, low cost, low-flying under clouds and ability to work full-time. When an earthquake happened, it could go deep into the places safely and reliably which human staff can hardly approach, such as secondary geological disasters hit areas. The system can be timely precise in response to secondary geological disasters monitoring by a way of obtaining first-hand information as quickly as possible, producing a unique emergency response capacity to provide a scientific basis for overall decision-making processes. It can greatly enhance the capability of on-site disaster emergency working team in data collection and transmission. The great advantages of UAV remote sensing system played an irreplaceable role in monitoring secondary geological disaster dynamics and influences. Taking the landslides and barrier lakes for example, the paper explored the basic application and process of UAV remote sensing in the disaster emergency relief. UAV high-resolution remote sensing images had been exploited to estimate the situation of disaster-hit areas and monitor secondary geological disasters rapidly, systematically and continuously. Furthermore, a rapid quantitative assessment on the distribution and size of landslides and barrier lakes was carried out. Monitoring results could support relevant government departments and rescue teams, providing detailed and reliable scientific evidence for disaster relief and decision-making.

In this study, the extended morphological attribute profiles (EAPs) and independent component analysis (ICA) were combined for feature extraction of high-resolution multispectral satellite remote sensing images and the regularized least squares (RLS) approach with the radial basis function (RBF) kernel was further applied for the classification. Based on the major two independent components, the geometrical features were extracted using the EAPs method. In this study, three morphological attributes were calculated and extracted for each independent component, including area, standard deviation, and moment of inertia. The extracted geometrical features classified results using RLS approach and the commonly used LIB-SVM library of support vector machines method. The Worldview-3 and Chinese GF-2 multispectral images were tested, and the results showed that the features extracted by EAPs and ICA can effectively improve the accuracy of the high-resolution multispectral image classification, ~2% larger than EAPs and principal component analysis (PCA) method, and ~6% larger than APs and original high-resolution multispectral data. Moreover, it is also suggested that both the GURLS and LIB-SVM libraries are well suited for the multispectral remote sensing image classification. The GURLS library is easy to be used with automatic parameter selection but its computation time may be larger than the LIB-SVM library. This study would be helpful for the classification application of high-resolution multispectral satellite remote sensing images.

Landscape structures and process on different scale show different characteristics. In the study of specific target landmarks, the most appropriate scale for images can be attained by scale conversion, which improves the accuracy and efficiency of feature identification and classification. In this paper, the authors carried out experiments on multi-scale classification by taking the Shangri-la area in the north-western Yunnan province as the research area and the images from SPOT5 HRG and GF-1 Satellite as date sources. Firstly, the authors upscaled the two images by cubic convolution, and calculated the optimal scale for different objects on the earth shown in images by variation functions. Then the authors conducted multi-scale superposition classification on it by Maximum Likelyhood, and evaluated the classification accuracy. The results indicates that: (1) for most of the object on the earth, the optimal scale appears in the bigger scale instead of the original one. To be specific, water has the biggest optimal scale, i.e. around 25-30m; farmland, grassland, brushwood, roads, settlement places and woodland follows with 20-24m. The optimal scale for shades and flood land is basically as the same as the original one, i.e. 8m and 10m respectively. (2) Regarding the classification of the multi-scale superposed images, the overall accuracy of the ones from SPOT5 HRG and GF-1 Satellite is 12.84% and 14.76% higher than that of the original multi-spectral images, respectively, and Kappa coefficient is 0.1306 and 0.1419 higher, respectively. Hence, the multi-scale superposition classification which was applied in the research area can enhance the classification accuracy of remote sensing images .

Digital watermarking has been recognized as a useful technology for the copyright protection and authentication of digital information. However, rarely did the former methods focus on the key content of digital carrier. The idea based on the protection of key content is more targeted and can be considered in different digital information, including text, image and video. In this paper, we use text as research object and a text zero-watermarking method which uses keyword dense interval (KDI) as the key content is proposed. First, we construct zero-watermarking model by introducing the concept of KDI and giving the method of KDI extraction. Second, we design detection model which includes secondary generation of zero-watermark and the similarity computing method of keyword distribution. Besides, experiments are carried out, and the results show that the proposed method gives better performance than other available methods especially in the attacks of sentence transformation and synonyms substitution.

The watermarking technique can be used to protect the copyright of relational databases by hiding the ownership information into the relational databases. Difference expansion (DE) technique is one of the common reversible watermarking techniques for numerical relational databases. However, most previous schemes based on DE suffer the problem of low embedding capacity when the difference values between different attributes are relatively large. In this paper, we propose a novel reversible watermarking scheme to solve the above problem. In the scheme, a mapping difference expansion (MDE) method is proposed to convert the differences between attributes to small mapping differences. Based on the MDE, an attribute and tuple selection algorithm is designed to select the suitable data for watermarking, which can increase embedding capacity and reduce distortion. In addition, the majority voting technique is utilized to enhance the robustness of watermarking with the high embedding capacity. The experimental results have shown that the proposed scheme can provide higher embedding capacity, lower distortion and stronger robustness than other schemes.

Digital Watermarking is a process of embedded data for copyright protection and authentication. This paper proposed a digitally invisible watermarking algorithm in video sequence by means of Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT). To improve the security, we used chaotic map for embedding position of a watermark. A watermark, binary image, was embedded in a selected sub-band. The experimental results show this algorithm is more robustness to attacks and can achieved copyright protection/authentication.

The reliable watermarking algorithm has been optimized by DD-PCA (Double-Decomposition Principal Component Analysis) using the relations and characteristics of PCA and SVD (Singular Value Decomposition). The principal component of the watermark is extracted using SVD and embedded into the singular values of the cover image. Then the second decomposition is used creatively. Experimental results show the algorithm proposed is greatly imperceptible-the PSNR can be over 40dp though embedding factor is 40. The algorithm is robust to noising, compression, blurring and geometric attacks. Besides, experimental results also show high reliability in solving the false positive.

We present a spatial domain image watermarking for printed and scanned documents. In the watermark embedding process, a watermark image is embedded into the blue color component of a white color image. The result is overlaid with information, and printed out on a piece of paper. In the watermark extraction process, a printed document is first scanned back to obtain an electronic copy. Geometric distortions from the printing and scanning processes are then reduced by an image registration technique based on affine transformation. All watermarked components are used to determine a threshold for watermark bits extraction. The performance of the proposed watermarking method was investigated based on different scanning resolutions, printing quality modes, and printable materials. Watermark extraction from ripped, crumpled, and wet documents was also investigated. The promising results demonstrate the effectiveness of the proposed method.

Fog effects in various computer and camera software can partially or fully damage the watermark information within the watermarked image. In this paper, we propose a color image watermarking based on the modification of reflectance component against fog effects. The reflectance component is extracted from the blue color channel in the RGB color space of a host image, and then used to carry a watermark signal. The watermark extraction is blindly achieved by subtracting the estimation of the original reflectance component from the watermarked component. The performance of the proposed watermarking method in terms of wPSNR and NC is evaluated, and then compared with the previous method. The experimental results on robustness against various levels of fog effect, from both computer software and mobile application, demonstrated a higher robustness of our proposed method, compared to the previous one.

In this paper, we propose a method for target detection and tracking in infrared video. The target is defined by its location and extent in a single frame. In the initialization process, we use an adaptive threshold to segment the target and then extract the fern feature and normalize it as a template. The detector uses the random forest and fern to detect the target in the infrared video. The random forest and fern is a random combination of 2bit Binary Pattern, which is robust to infrared targets with blurred and unknown contours. The tracker uses the gray-value weighted mean-Shift algorithm to track the infrared target which is always brighter than the background. And the tracker can track the deformed target efficiently and quickly. When the target disappears, the detector will redetect the target in the coming infrared image. Finally, we verify the algorithm on the real-time infrared target detection and tracking platform. The result shows that our algorithm performs better than TLD in terms of recall and runtime in infrared video.

Moving targets detection is gaining popularity in civilian and military applications. On some monitoring platform of motion detection, some low-resolution stationary cameras are replaced by moving HD camera based on UAVs. The pixels of moving targets in the HD Video taken by UAV are always in a minority, and the background of the frame is usually moving because of the motion of UAVs. The high computational cost of the algorithm prevents running it at higher resolutions the pixels of frame. Hence, to solve the problem of moving targets detection based UAVs video, we propose a heterogeneous CPU-GPU moving target detection algorithm for UAV video. More specifically, we use background registration to eliminate the impact of the moving background and frame difference to detect small moving targets. In order to achieve the effect of real-time processing, we design the solution of heterogeneous CPU-GPU framework for our method. The experimental results show that our method can detect the main moving targets from the HD video taken by UAV, and the average process time is 52.16ms per frame which is fast enough to solve the problem.

This paper proposes a no-reference objective stereoscopic video quality assessment method with the motivation that making the effect of objective experiments close to that of subjective way. We believe that the image regions with different visual salient degree should not have the same weights when designing an assessment metric. Therefore, we firstly use GBVS algorithm to each frame pairs and separate both the left and right viewing images into the regions with strong, general and week saliency. Besides, local feature information like blockiness, zero-crossing and depth are extracted and combined with a mathematical model to calculate a quality assessment score. Regions with different salient degree are assigned with different weights in the mathematical model. Experiment results demonstrate the superiority of our method compared with the existed state-of-the-art no-reference objective Stereoscopic video quality assessment methods.

Image and video dehazing is a popular topic in the field of computer vision and digital image processing. A fast, optimized dehazing algorithm was recently proposed that enhances contrast and reduces flickering artifacts in a dehazed video sequence by minimizing a cost function that makes transmission values spatially and temporally coherent. However, its fixed-size block partitioning leads to block effects. Further, the weak edges in a hazy image are not addressed. Hence, a video dehazing algorithm based on customized spectral clustering is proposed. To avoid block artifacts, the spectral clustering is customized to segment static scenes to ensure the same target has the same transmission value. Assuming that dehazed edge images have richer detail than before restoration, an edge cost function is added to the ransmission model. The experimental results demonstrate that the proposed method provides higher dehazing quality and lower time complexity than the previous technique.

Automatic detection of abnormal cells from cervical smear images is extremely demanded in annual diagnosis of women’s cervical cancer. For this medical cell recognition problem, there are three different feature sections, namely cytology morphology, nuclear chromatin pathology and region intensity. The challenges of this problem come from feature combination s and classification accurately and efficiently. Thus, we propose an efficient abnormal cervical cell detection system based on multi-instance extreme learning machine (MI-ELM) to deal with above two questions in one unified framework. MI-ELM is one of the most promising supervised learning classifiers which can deal with several feature sections and realistic classification problems analytically. Experiment results over Herlev dataset demonstrate that the proposed method outperforms three traditional methods for two-class classification in terms of well accuracy and less time.

Age-related Macular Degeneration (AMD) is a kind of macular disease which mostly occurs in old people，and it may cause decreased vision or even lead to permanent blindness. Drusen is an important clinical indicator for AMD which can help doctor diagnose disease and decide the strategy of treatment. Optical Coherence Tomography (OCT) is widely used in the diagnosis of ophthalmic diseases, include AMD. In this paper, we propose a classification method based on Multiple Instance Learning (MIL) to detect AMD. Drusen can exist in a few slices of OCT images, and MIL is utilized in our method. We divided the method into two phases: training phase and testing phase. We train the initial features and clustered to create a codebook, and employ the trained classifier in the test set. Experiment results show that our method achieved high accuracy and effectiveness.

The morphological change of red blood cells(RBCs) plays an important role in revealing the biomechanical and biorheological characteristics of RBCs. Aiming to extract the shape indices for the sickle RBCs, an automated ex-vivo RBC shape quantification method is proposed. First, single RBC regions (ROIs) are extracted from raw microscopy image via an automatic hierarchical ROI extraction method. Second, an improved random walk method is used to detect the RBC outline. Finally, three types of RBC shape factors are calculated based on the elliptical fitting RBC contour. Experiments indicate that the proposed method can accurately segment the RBCs from the microscopy images with low contrast and prevent the disturbance of artifacts. Moreover, it can provide an efficient shape quantification means for diverse RBC shapes in a batch manner.

The study takes advantage of several new breakthroughs in computer vision technology to develop a new mid-irisbiomedical platform that processes iris image for early detection of heart-disease. Guaranteeing early detection of heart disease provides a possibility of having non-surgical treatment as suggested by biomedical researchers and associated institutions. However, our observation discovered that, a clinical practicable solution which could be both sensible and specific for early detection is still lacking. Due to this, the rate of majority vulnerable to death is highly increasing. The delayed diagnostic procedures, inefficiency, and complications of available methods are the other reasons for this catastrophe. Therefore, this research proposes the novel IFB (Iris Features Based) method for diagnosis of premature, and early stage heart disease. The method incorporates computer vision and iridology to obtain a robust, non-contact, nonradioactive, and cost-effective diagnostic tool. The method analyzes abnormal inherent weakness in tissues, change in color and patterns, of a specific region of iris that responds to impulses of heart organ as per Bernard Jensen-iris Chart. The changes in iris infer the presence of degenerative abnormalities in heart organ. These changes are precisely detected and analyzed by IFB method that includes, tensor-based-gradient(TBG), multi orientations gabor filters(GF), textural oriented features(TOF), and speed-up robust features(SURF). Kernel and Multi class oriented support vector machines classifiers are used for classifying normal and pathological iris features. Experimental results demonstrated that the proposed method, not only has better diagnostic performance, but also provides an insight for early detection of other diseases.

Endoscopy imaging is utilized to provide a realistic view about the surfaces of organs inside the human body. Owing to the damp internal environment, these surfaces usually have a glossy appearance showing specular reflections. For many computer vision algorithms, the highlights created by specular reflections may become a significant source of error. In this paper, we present a novel method for restoration of the specular reflection regions from a single image. Specular restoration process starts with generating a substitute specular-free image with RPCA method. Then the specular removed image was obtained by taking the binary weighting template of highlight regions as the weighting for merging the original specular image and the substitute image. The modified template was furthermore discussed for the concealment of artificial effects in the edge of specular regions. Experimental results on the removal of the endoscopic image with specular reflections demonstrate the efficiency of the proposed method comparing to the existing methods.

Aiming at solving the defect, slow speed of magnetic resonance imaging (MRI), we propose a magnetic resonance image reconstruction method based on compressed sensing (CS), combining Shift-Invariant Discrete Wavelet Transformation (SIDWT) and dictionary learning (DL). Since the method is based on CS, it obtains the part of k-space image data by undersampling and reconstructs the original image by SIDWT. Then the dictionary learning and sparse coding phase are used in the sparse representation of the original image. Finally, the undersampling image is reconstructed by image iteration update stage, and the quality of the reconstructed image is evaluated objectively via the Peak Signal-to-Noise Ratio (PSNR). Compared with the existing sparse reconstruction algorithm, the proposed method achieves high quality reconstruction of undersampled MR images, and shortens the time of MR image signal acquisition.

Human Epithelial-2 (HEp-2) cell images staining patterns classification have been widely used to identify autoimmune diseases by the anti-Nuclear antibodies (ANA) test in the Indirect Immunofluorescence (IIF) protocol. Because manual test is time consuming, subjective and labor intensive, image-based Computer Aided Diagnosis (CAD) systems for HEp-2 cell classification are developing. However, methods proposed recently are mostly manual features extraction with low accuracy. Besides, the scale of available benchmark datasets is small, which does not exactly suitable for using deep learning methods. This issue will influence the accuracy of cell classification directly even after data augmentation. To address these issues, this paper presents a high accuracy automatic HEp-2 cell classification method with small datasets, by utilizing very deep convolutional networks (VGGNet). Specifically, the proposed method consists of three main phases, namely image preprocessing, feature extraction and classification. Moreover, an improved VGGNet is presented to address the challenges of small-scale datasets. Experimental results over two benchmark datasets demonstrate that the proposed method achieves superior performance in terms of accuracy compared with existing methods.

Thyroid Imaging Reporting and Data System(TI-RADS) is a valuable tool for differentiating the benign and the malignant thyroid nodules. In clinic, doctors can determine the extent of being benign or malignant in terms of different classes by using TI-RADS. Classification represents the degree of malignancy of thyroid nodules. TI-RADS as a classification standard can be used to guide the ultrasonic doctor to examine thyroid nodules more accurately and reliably. In this paper, we aim to classify the thyroid nodules with the help of TI-RADS. To this end, four ultrasound signs, i.e., cystic and solid, echo pattern, boundary feature and calcification of thyroid nodules are extracted and converted into feature vectors. Then semi-supervised fuzzy C-means ensemble (SS-FCME) model is applied to obtain the classification results. The experimental results demonstrate that the proposed method can help doctors diagnose the thyroid nodules effectively.