Color harmonization is an artistic technique to adjust a set of colors in order to enhance their visual harmony so that they are aesthetically pleasing in terms of human visual perception. We present a new color harmonization method that treats the harmonization as a function optimization. For a given image, we derive a cost function based on the observation that pixels in a small window that have similar unharmonic hues should be harmonized with similar harmonic hues. By minimizing the cost function, we get a harmonized image in which the spatial coherence is preserved. A new matching function is proposed to select the best matching harmonic schemes, and a new component-based preharmonization strategy is proposed to preserve the hue distribution of the harmonized images. Our approach overcomes several shortcomings of the existing color harmonization methods. We test our algorithm with a variety of images to demonstrate the effectiveness of our approach.

The purpose of this study is to develop a three-dimensional registration method for monitoring knee joint disease from magnetic resonance (MR) image data sets. A global optimization technique was used for identifying anatomically corresponding points of knee femur surfaces (bone cartilage interfaces). In a first pre-registration step, we used the principal axes transformation to correct for different knee joint positions and orientations in the MR scanner. In a second step, we presented a global search algorithm based on Lipschitz optimization theory. This technique can simultaneously determine the translation and rotation parameters through searching a six-dimensional space of Euclidean motion metrics (translation and rotation) after calculating the point correspondences. The point correspondences were calculated by using the Hungarian algorithm. The accuracy of registration was evaluated using 20 porcine knees. There were 300 corresponding landmark points over the 20 pig knees. We evaluated the registration accuracy by measuring the root-mean-square distance (RMSD) error of corresponding landmark points between two femur surfaces (two time-points). The results show that the average RMSD was 1.22 ± 0.10 mm (SD) by the iterative closest point (ICP) method, 1.17 ± 0.10 mm the by expectation-maximization-ICP method, 1.02 ± 0.06 mm by the genetic method, and 0.93 ± 0.04 mm by the proposed method. Compared with the other three registration approaches, the proposed method achieved the highest registration accuracy.

In this paper, we present a multispectral image (MSI) compression method using a lossless and lossy coding scheme, which focuses on the seamless coding of the RGB bit stream to enhance the usability of the MSI. The proposed method divides the MSI data into two components: RGB and residual. The RGB component is extracted from the MSI by using the XYZ color matching functions, a color conversion matrix, and a gamma curve. The original MSI is estimated by an RGB data encoder and the difference between the original and the estimated MSI, which is referred to as the residual component in this paper. Next, the RGB and residual components are encoded by using JPEG2000, and progressive decoding is achieved from the losslessly encoded code stream. Experimental results show that a high-quality RGB image can be obtained at a low bit rate with primary encoding of the RGB component. In addition, by using the proposed method, the quality of a spectrum can be improved by decoding the residual data, and the quality is comparable to that obtained by using JPEG2000. The lossless compression ratio obtained by using this method is also similar to that obtained by using JPEG2000 with the integer Karhunen-Loeve transform.

We propose a novel variational formulation (external energy) for implicit active contours that forces the level-set function to have the opposite sign along the edges at convergence. This external energy is then incorporated into a variational level-set formulation with two extra regularization terms (internal energy). The resulting evolution of the level-set function is the gradient flow that minimizes the overall energy functional. Because of the external energy, the level-set function can be initialized to any bounded function (e.g., a constant function), which completely eliminates the need of initial contours. This implies that the new formulation is robust to initialization or even free of manual initialization. The proposed model is been applied to both real and synthetic images with promising results. Besides, our model is especially appropriate for segmentation of real images with a complex and /or nonuniform background.

Learning-based methods are well adopted in image super-resolution. In this paper, we propose a new learning-based approach using contourlet transform and Markov random field. The proposed algorithm employs contourlet transform rather than the conventional wavelet to represent image features and takes into account the correlation between adjacent pixels or image patches through the Markov random field (MRF) model. The input low-resolution (LR) image is decomposed with the contourlet transform and fed to the MRF model together with the contourlet transform coefficients from the low- and high-resolution image pairs in the training set. The unknown high-frequency components/coefficients for the input low-resolution image are inferred by a belief propagation algorithm. Finally, the inverse contourlet transform converts the LR input and the inferred high-frequency coefficients into the super-resolved image. The effectiveness of the proposed method is demonstrated with the experiments on facial, vehicle plate, and real scene images. A better visual quality is achieved in terms of peak signal to noise ratio and the image structural similarity measurement.

A zoom lens calibration consists of hundreds of monofocal calibrations, each of which takes considerable time and effort with conventional methods. We present a practical calibration method that consists of two separate procedures-zoom calibration and focus calibration. The zoom calibration regards each zoom setting as a monofocal camera, and takes advantage of both the pattern-based and the rotation-based approaches. A rotation sensor is utilized to overcome the ill-posedness caused by the many parameter dimensions. The zoom calibration is followed by the focus calibration process, which is fully automatic and available even at a defocused setting where the pattern detection is not possible. The focus calibration drastically reduces the number of required manual calibrations, from N² to N times. The experimental results are compared with the lens data sheets provided by the lens manufacturer. The overall calibration procedure is very quick compared to the conventional methods, owing to the proposed zoom and focus calibrations, and shows small enough parameter errors in the effective zoom-focus range.

We present high-performance and low-complexity algorithms for real-time camera imaging applications. The main functions of the proposed camera digital signal processing (DSP) involve color interpolation, white balance, adaptive binary processing, auto gain control, and edge and color enhancement for video projection systems. A series of simulations demonstrate that the proposed method can achieve good image quality while keeping computation cost and memory requirements low. On the basis of the proposed algorithms, the cost-effective hardware core is developed using Verilog HDL. The prototype chip has been verified with one low-cost programmable device. The real-time camera system can achieve 1270 × 792 resolution with the combination of extra components and can demonstrate each DSP function.

Colorization is the process of adding colors to monochrome images. State-of-the-art colorization methods can be generally categorized into example-based colorization and scribble-based algorithms. In this paper, we present a new scribble-based colorization algorithm based on Bayesian inference and nonlocal likelihood computation. We convert the process of image colorization to a probability optimization problem in this Bayesian framework, where we use nonlocal-mean likelihood computation and Markov random field prior's. The expectation maximization method is used to solve an optimization object function. Finally, experimental results demonstrate the effectiveness of the proposed algorithm

When a digital image sequence is sampled with a periodically emitting light source, the color rolling (CR) phenomenon occurs, which is shown by periodical variations of color and luminance values. In conventional CR suppression (CRS) methods, color variation has been reduced by using auto white balance methods. However, the CR phenomenon still appears in the resulting image sequences due to interfield illuminant intensity variation. In the proposed CRS method, the interfield luminance and color variations are simultaneously suppressed by estimating the illuminant change between the current and the target fields. In order to consider the object motions, a motion detection technique is used to estimate the luminance changes that occurred due to the CR phenomenon. Moreover, the illuminant color is estimated using the CR achromatic color distribution in the chromaticity space which is founded on the periodicity of the CR phenomenon. Based on the motion detection and the achromatic color detection techniques, the illuminant is estimated using the obtained color components in a common area of both static and achromatic regions. The experimental results demonstrate that our strategy efficiently suppresses the CR phenomenon without being affected by moving objects and produces luminance and color constant image sequences.

H.264/AVC is the newest standard for digital video compression developed jointly by ITU-T's Video Coding Experts Group and ISO/IEC's Moving Picture Experts Group. One feature of the new standard is the adoption of a robust error resilience tool at the encoder known as flexible macroblock ordering (FMO). In this paper, we present an algorithm to generate a one-pass FMO map based on spatial and temporal information at the encoder, and an error concealment method at the decoder for wireless video transmission. The error concealment method at the decoder is applied according to the residual information derived from the distortion information obtained at the encoder while the one-pass FMO map is being generated. Our simulation results performed under slow and fast fading channels confirm that the proposed technique can reduce the number of undecodable macroblocks up to 66.79% and 80.54%, respectively, when compared with no FMO. The peak signal-to-noise ratio improvements are up to 2.67 and 1.05 dB, respectively when compared to predefined FMO (Type 1).

Digital time delay and integration (DTDI) mosaic video sequences captured by high-speed DTDI line-scan cameras are commonly used in industrial print inspection and high-speed capture applications. To reduce the memory requirement for saving these video sequences, it is necessary to compress them. In this paper, we present an efficient chroma subsampling strategy for compressing DTDI mosaic video sequences in H.264/AVC. Based on the color domain transform between the RGB domain and the YUV domain, a position-selection strategy is proposed to determine the two subsampling chroma components, U and V, according to the DTDI mosaic structure. The quality of reconstructed DTDI video sequences is better than those reconstructed by conventional methods. By experimenting on some popular test DTDI mosaic video sequences, the results turned out to be superior than conventional ones that adopt H.264/AVC as compression standard.

It has long been known that there are numerous advantages to sampling images hexagonally rather than rectangularly. However, due to various shortcomings of the addressing schemes, hexagonal sampling for digital images has not been embraced by the mainstream digital imaging community. The idea of using hexagonal sampling for digital imaging applications has been around since the early 1960s, yet no efficient addressing method for hexagonal grids has been developed in that time. This paper introduces a new hexagonal addressing approach, called array set addressing (ASA), that solves the problems exhibited by other addressing methods. The ASA approach uses three coordinates to represent the hexagonal grid as a pair of rectangular arrays. This representation supports efficient linear algebra and image processing manipulation. ASA-based implementations of several basic image processing operations are presented and shown to be efficient. A hexagonal fast Fourier transform, based on the fact that the Fourier kernel becomes separable when using ASA coordinates, is also presented.

Image super-resolution aims to obtain a high-quality image at a resolution that is higher than that of the original coarse one. This paper presents a new neural network-based method for image super-resolution. In this technique, the super-resolution is considered as an inverse problem. An observation model that closely follows the physical image acquisition process is established to solve the problem. Based on this model, a cost function is created and minimized by a Hopfield neural network to produce high-resolution images from the corresponding low-resolution ones. Not like some other single frame super-resolution techniques, this technique takes into consideration point spread function blurring as well as additive noise and therefore generates high-resolution images with more preserved or restored image details. Experimental results demonstrate that the high-resolution images obtained by this technique have a very high quality in terms of PSNR and visually look more pleasant.

This paper presents the design of the SCAN secure processor and its extended instruction set to enable secure biometric authentication. The SCAN secure processor is a modified SparcV8 processor architecture with a new instruction set to handle voice, iris, and fingerprint-based biometric authentication. The algorithms for processing biometric data are based on the local global graph methodology. The biometric modules are synthesized in reconfigurable logic and the results of the field-programmable gate array (FPGA) synthesis are presented. We propose to implement the above-mentioned modules in an off-chip FPGA co-processor. Further, the SCAN-secure processor will offer a SCAN-based encryption and decryption of 32 bit instructions and data.

A video sequence consists of several hundred frames, and as a result, creating a panoramic image from these frames is a very time-consuming process. Consecutive frames have large overlap areas that do not provide much information. Therefore, some key frames must be extracted for better performance. There are a number of methods for key-frame selection that match all frames in a video sequence. We present a novel and efficient method to select key frames from video for creating a large panoramic mosaic without matching all frames. Consecutive frames are transformed and projected onto the common mosaic surface and the position of each corner of the next frame is predicted with a distinct Kalman filter on this surface. The overlap area between each predicted frame and its previous key frame is used as the criterion to select the next key frame. This method uses video information to reduce features and align frames with repeated structures more accurately. We show that this approach is an efficient preprocessing step and substantially reduces the time required to construct panorama from a video sequence.

Single sensor digital color cameras capture only one of the three primary colors at each pixel and a process called color demosaicking (CDM) is used to reconstruct the full color images. Most CDM algorithms assume the existence of high local spectral redundancy in estimating the missing color samples. However, for images with sharp color transitions and high color saturation, such an assumption may be invalid and visually unpleasant CDM errors will occur. In this paper, we exploit the image nonlocal redundancy to improve the local color reproduction result. First, multiple local directional estimates of a missing color sample are computed and fused according to local gradients. Then, nonlocal pixels similar to the estimated pixel are searched to enhance the local estimate. An adaptive thresholding method rather than the commonly used nonlocal means filtering is proposed to improve the local estimate. This allows the final reconstruction to be performed at the structural level as opposed to the pixel level. Experimental results demonstrate that the proposed local directional interpolation and nonlocal adaptive thresholding method outperforms many state-of-the-art CDM methods in reconstructing the edges and reducing color interpolation artifacts, leading to higher visual quality of reproduced color images.

Blind source separation is a process in which mixed signals, obtained as a linear combination of various source signals, are decomposed into their original sources. The source signals and their mixture weights are unknown, but a priori information about their statistical behavior and mixing model is available. In this paper, a new algorithm based on generalized cross correlation linear-operator set is proposed. This algorithm significantly improves source-separation quality compared to several other well-known algorithms, such as subband decomposition independent component analysis, block Gaussian likelihood, and convex analysis of mixtures of non-negative sources.

A multiscale operator for spatiotemporal isotropic attention is proposed to reliably extract attention points during image sequence analysis. Its consecutive local maxima indicate attention points as the centers of image fragments of variable size with high intensity contrast, region homogeneity, regional shape saliency, and temporal change presence. The scale-adaptive estimation of temporal change (motion) and its aggregation with the regional shape saliency contribute to the accurate determination of attention points in image sequences. Multilocation descriptors of an image sequence are extracted at the attention points in the form of a set of multidimensional descriptor vectors. A fast recursive implementation is also proposed to make the operator's computational complexity independent from the spatial scale size, which is the window size in the spatial averaging filter. Experiments on the accuracy of attention-point detection have proved the operator consistency and its high potential for multiscale feature extraction from image sequences.

A book review of a recently released title.