Search

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.

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.

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.

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.

We propose a new concept of tuning a point-spread function (a “kernel” function) in the modeling of electron beam lithography using the machine learning scheme. Normally in the work of artificial intelligence, the researchers focus on the output results from a neural network, such as success ratio in image recognition or improved production yield, etc. In this work, we put more focus on the weights connecting the nodes in a convolutional neural network, which are naturally the fractions of a point-spread function, and take out those weighted fractions after learning to be utilized as a tuned kernel. Proof-of-concept of the kernel tuning has been demonstrated using the examples of proximity effect correction with 2-layer network, and charging effect correction with 3-layer network. This type of new tuning method can be beneficial to give researchers more insights to come up with a better model, yet it might be too early to be deployed to production to give better critical dimension (CD) and positional accuracy almost instantly.

Algorithms for radar signal processing, such as Synthetic Aperture Radar (SAR) are computationally intensive and require considerable execution time on a general purpose processor. Reconfigurable logic can be used to off-load the primary computational kernel onto a custom computing machine in order to reduce execution time by an order of magnitude as compared to kernel execution on a general purpose processor. Specifically, Field Programmable Gate Arrays (FPGAs) can be used to accelerate these kernels using hardware-based custom logic implementations. In this paper, we demonstrate a framework for algorithm acceleration. We used SAR as a case study to illustrate the potential for algorithm acceleration offered by FPGAs. Initially, we profiled the SAR algorithm and implemented a homomorphic filter using a hardware implementation of the natural logarithm. Experimental results show a linear speedup by adding reasonably small processing elements in Field Programmable Gate Array (FPGA) as opposed to using a software implementation running on a typical general purpose processor.

An iterative algorithm is proposed for blind multi-image deblurring of binary images. The binarity is the only prior restriction imposed on the image. Image formation model assumes convolution with arbitrary kernel and addition of a constant value. Penalty functional is composed using binarity constraint for regularization. The algorithm estimates the original image and distortion parameters by alternate reduction of two parts of this functional. Experimental results for natural (non-synthetic) data are present.

Freehand three-dimensional (3D) ultrasound imaging enables low-cost and flexible 3D scanning of arbitrary-shaped organs, where the operator can freely move a two-dimensional (2D) ultrasound probe to acquire a sequence of tracked cross-sectional images of the anatomy. Often, the acquired 2D ultrasound images are irregularly and sparsely distributed in the 3D space. Several 3D reconstruction algorithms have been proposed to synthesize 3D ultrasound volumes based on the acquired 2D images. A challenging task during the reconstruction process is to preserve the texture patterns in the synthesized volume and ensure that all gaps in the volume are correctly filled. This paper presents an adaptive kernel regression algorithm that can effectively reconstruct high-quality freehand 3D ultrasound volumes. The algorithm employs a kernel regression model that enables nonparametric interpolation of the voxel gray-level values. The kernel size of the regression model is adaptively adjusted based on the characteristics of the voxel that is being interpolated. In particular, when the algorithm is employed to interpolate a voxel located in a region with dense ultrasound data samples, the size of the kernel is reduced to preserve the texture patterns. On the other hand, the size of the kernel is increased in areas that include large gaps to enable effective gap filling. The performance of the proposed algorithm was compared with seven previous interpolation approaches by synthesizing freehand 3D ultrasound volumes of a benign breast tumor. The experimental results show that the proposed algorithm outperforms the other interpolation approaches.

Positron emission tomography (PET) imaging has been widely explored for treatment outcome prediction. Radiomicsdriven methods provide a new insight to quantitatively explore underlying information from PET images. However, it is still a challenging problem to automatically extract clinically meaningful features for prognosis. In this work, we develop a PET-guided distant failure predictive model for early stage non-small cell lung cancer (NSCLC) patients after stereotactic ablative radiotherapy (SABR) by using sparse representation. The proposed method does not need precalculated features and can learn intrinsically distinctive features contributing to classification of patients with distant failure. The proposed framework includes two main parts: 1) intra-tumor heterogeneity description; and 2) dictionary pair learning based sparse representation. Tumor heterogeneity is initially captured through anisotropic kernel and represented as a set of concatenated vectors, which forms the sample gallery. Then, given a test tumor image, its identity (i.e., distant failure or not) is classified by applying the dictionary pair learning based sparse representation. We evaluate the proposed approach on 48 NSCLC patients treated by SABR at our institute. Experimental results show that the proposed approach can achieve an area under the characteristic curve (AUC) of 0.70 with a sensitivity of 69.87% and a specificity of 69.51% using a five-fold cross validation.

Image reconstruction in diffuse optical tomography (DOT) is challenging because its inverse problem is nonlinear, ill-posed and ill-conditioned. Anatomical guidance from high spatial resolution imaging modalities can substantially improve the quality of reconstructed DOT images. In this paper, inspired by the kernel methods in machine learning, we propose the kernel method to introduce anatomical information into the DOT image reconstruction algorithm. In this kernel method, optical absorption coefficient at each finite element node is represented as a function of a set of features obtained from anatomical images such as computed tomography (CT). The kernel based image model is directly incorporated into the forward model of DOT, which exploits the sparseness of the image in the feature space. Compared with Laplacian approaches to include structural priors, the proposed method does not require the image segmentation of distinct regions. The proposed kernel method is validated with numerical simulations of 3D DOT reconstruction using synthetic CT data. We added 15% Gaussian noise onto both the numerical DOT measurements and the simulated CT image. We have also validated the proposed method by agar phantom experiment with anatomical guidance from a CT scan. We have studied the effects of voxel size and number of nearest neighborhood size in kernel method on the reconstructed DOT images. Our results indicate that the spatial resolution and the accuracy of the reconstructed DOT images have been improved substantially after applying the anatomical guidance with the proposed kernel method.

Reconstruction of fluorescence molecular tomography (FMT) is an ill-posed inverse problem. Anatomical guidance in the FMT reconstruction can improve FMT reconstruction efficiently. We have developed a kernel method to introduce the anatomical guidance into FMT robustly and easily. The kernel method is from machine learning for pattern analysis and is an efficient way to represent anatomical features. For the finite element method based FMT reconstruction, we calculate a kernel function for each finite element node from an anatomical image, such as a micro-CT image. Then the fluorophore concentration at each node is represented by a kernel coefficient vector and the corresponding kernel function. In the FMT forward model, we have a new system matrix by multiplying the sensitivity matrix with the kernel matrix. Thus, the kernel coefficient vector is the unknown to be reconstructed following a standard iterative reconstruction process. We convert the FMT reconstruction problem into the kernel coefficient reconstruction problem. The desired fluorophore concentration at each node can be calculated accordingly. Numerical simulation studies have demonstrated that the proposed kernel-based algorithm can improve the spatial resolution of the reconstructed FMT images. In the proposed kernel method, the anatomical guidance can be obtained directly from the anatomical image and is included in the forward modeling. One of the advantages is that we do not need to segment the anatomical image for the targets and background.

The population is aging as the years pass. There is an increase in life expectancy, but also a decrease in the quality of life for the presence of chronic degenerative diseases. Processing medical images can identify brain changes typical of Alzheimer's disease (AD) and Mild Cognitive Impairment (MCI) at an early stage. We propose a new method of segmentation technique using Mean Shift algorithm applying probabilistic maps and Support Vector Machine with Linear and Radial Basis Kernel for segmentation of the hippocampus on Magnetic Resonance Images (MRI). The similarity index of DICE for a 8 control subject was calculated obtaining a mean value of 0:7053±0:0996 using Linear kernel and 0:7275±0:1335 using RBF kernel compared with the manual segmentation made by radiologist.

Three-dimensional measurement is the base part for reverse engineering. The paper developed a new flexible and fast optical measurement method based on multi-view geometry theory. At first, feature points are detected and matched with improved SIFT algorithm. The Hellinger Kernel is used to estimate the histogram distance instead of traditional Euclidean distance, which is immunity to the weak texture image; then a new filter three-principle for filtering the calculation of essential matrix is designed, the essential matrix is calculated using the improved a Contrario Ransac filter method. One view point cloud is constructed accurately with two view images; after this, the overlapped features are used to eliminate the accumulated errors caused by added view images, which improved the camera’s position precision. At last, the method is verified with the application of dental restoration CAD/CAM, experiment results show that the proposed method is fast, accurate and flexible for tooth 3D measurement.

Recently, Kernel Correlation Filter (KCF) has achieved great attention in visual tracking filed, which provide excellent tracking performance and high possessing speed. However, how to handle the scale variation is still an open problem. In this paper, focusing on this issue that a method based on Gaussian scale space is proposed. First, we will use KCF to estimate the location of the target, the context region which includes the target and its surrounding background will be the image to be matched. In order to get the matching image of a Gaussian scale space, image with Gaussian kernel convolution can be gotten. After getting the Gaussian scale space of the image to be matched, then, according to it to estimate target image under different scales. Combine with the scale parameter of scale space, for each corresponding scale image performing bilinear interpolation operation to change the size to simulate target imaging at different scales. Finally, matching the template with different size of images with different scales, use Mean Absolute Difference (MAD) as the match criterion. After getting the optimal matching in the image with the template, we will get the best zoom ratio s, consequently estimate the target size. In the experiments, compare with CSK, KCF etc. demonstrate that the proposed method achieves high improvement in accuracy, is an efficient algorithm.

Shift-invariant motion blur can be modeled as a convolution of the true latent image and the blur kernel with additive noise. Blind motion de-blurring estimates a sharp image from a motion blurred image without the knowledge of the blur kernel. This paper proposes an improved edge-specific motion de-blurring algorithm which proved to be fit for processing remote sensing images. We find that an inaccurate blur kernel is the main factor to the low-quality restored images. To improve image quality, we do the following contributions. For the robust kernel estimation, first, we adapt the multi-scale scheme to make sure that the edge map could be constructed accurately; second, an effective salient edge selection method based on RTV (Relative Total Variation) is used to extract salient structure from texture; third, an alternative iterative method is introduced to perform kernel optimization, in this step, we adopt l1 and l0 norm as the priors to remove noise and ensure the continuity of blur kernel. For the final latent image reconstruction, an improved adaptive deconvolution algorithm based on TV-l2 model is used to recover the latent image; we control the regularization weight adaptively in different region according to the image local characteristics in order to preserve tiny details and eliminate noise and ringing artifacts. Some synthetic remote sensing images are used to test the proposed algorithm, and results demonstrate that the proposed algorithm obtains accurate blur kernel and achieves better de-blurring results.

This study investigates methods for characterizing materials that are mixtures of granular solids, or mixtures of liquids, which may be linear or non-linear. Linear mixtures of materials in a scene are often the result of areal mixing, where the pixel size of a sensor is relatively large so they contain patches of different materials within them. Non-linear mixtures are likely to occur with microscopic mixtures of solids, such as mixtures of powders, or mixtures of liquids, or wherever complex scattering of light occurs. This study considers two approaches for use as generalized methods for un-mixing pixels in a scene that may be linear or non-linear. One method is based on earlier studies that indicate non-linear mixtures in reflectance space are approximately linear in albedo space. This method converts reflectance to single-scattering albedo (SSA) according to Hapke theory assuming bidirectional scattering at nadir look angles and uses a constrained linear model on the computed albedo values. The other method is motivated by the same idea, but uses a kernel that seeks to capture the linear behavior of albedo in non-linear mixtures of materials. The behavior of the kernel method can be highly dependent on the value of a parameter, gamma, which provides flexibility for the kernel method to respond to both linear and non-linear phenomena. Our study pays particular attention to this parameter for responding to linear and non-linear mixtures. Laboratory experiments on both granular solids and liquid solutions are performed with scenes of hyperspectral data.

The paper presents principal approaches to diagnosing the structure forming skeleton of the complex optical field. An analysis of optical field singularity algorithms depending on intensity discretization and image resolution has been carried out. An optimal approach is chosen, which allows to bring much closer the solution of the phase problem of localization speckle-field special points. The use of a “window” 2D Hilbert transform for reconstruction of the phase distribution of the intensity of a speckle field is proposed. It is shown that the advantage of this approach consists in the invariance of a phase map to a change of the position of the kernel of transformation and in a possibility to reconstruct the structure-forming elements of the skeleton of an optical field, including singular points and saddle points. We demonstrate the possibility to reconstruct the equi-phase lines within a narrow confidence interval, and introduce an additional algorithm for solving the phase problem for random 2D intensity distributions.

This paper proposes a kind of fast image encryption algorithm based on permutation and diffusion architecture. An improved 1D chaotic map with three control parameters is adopted to enlarge the key space. Generalized Fibonacci- Lucas transform (GFLT) is utilized to change the positions of pixels of plain image in permutation. In order to enhance the security, the round number of permutation relates to image itself and the transform kernel of GFLT is varied and depends on both the chaotic map and plain image. In diffusion stage, the bidirectional diffusion operation is adopted, which reduces the time cost comparing with general diffusion process. Meanwhile, secure hash algorithm (SHA) is used to produce external key streams, which constitute the initial values of chaotic map and the round number of permutation. Hence, our encryption has large key space to resist the brute attack and the ability of resistance of the chosen/known attack due to the improved 1D chaotic map and SHA. Experimental simulations and security analysis both demonstrate that the proposed image encryption method can enhance the security level and at the same time reduce the computation load.

The aim of this paper was to design and implement an information technology (IT) system supporting the analysis and interpretation of image descriptors. The software is characterized by its versatility and speed in operating while processing series of digital images. The computer system can be expanded by new methods and is dedicated as a kernel of an expert system. The application seeks to extract the parameters of quality characteristics of agricultural crops - in this case, potatoes – in order to generate a set of data as a .csv file. The system helps to prepare the assessment of quality parameters of potatoes and generate mathematical models using Artificial Neural Network (ANN) simulators such as MATLAB ANN Toolbox or STATISTICA ANN toolbox in order to create a training dataset and information in that dataset.

Recovering a latent image from its blurry version is a severely ill-posed problem. In this paper a post process method is proposed for accurately estimating motion blur kernel based on its prior knowledge. And considering the small details destroy blur kernel estimation, an image decomposition process is executed before the estimation, which can decompose the image into cartoon and texture components. In the iterative framework, the cartoon that contains the main structures will be used to blur kernel estimation for avoiding the artifacts introduced by small texture. In addition, the algorithm adopted in the paper dynamically adjusts the size of patch which contains blur kernel instead of using the fix one as other works. Experimental results show that our method can get the more precise blur kernel and obtain the inspiring deblurring version from single blurry image.

One can process images acquired by single telescopes using adaptve optics (AO) in a manner similar to data usually acquired with a non-redundant aperture mask. Because it relies on a linearization of the phase transfer equation that describes the value of the phase in the Fourier-plane, the kernel- and eigen-phase analysis of images corrected by adaptive-optics is intrinsically limited to a regime of aberration where the rms error is less than the wavelength. This paper outlines the properties of an extension of the Fourier-phase analysis method, here applied to a set of spectrally dispersed data, similar to those produced by an intergral field spectrograph. Given sufficient wavelength coverage, the original baseline mapping model can be used to produce novel spectrally dispersed kernel-phase information for any aberration regime.

The QIS was conceived when contemplating shrinking pixel sizes and storage capacities, and the steady increase in digital processing power. In the single-bit QIS, the output of each field is a binary bit plane, where each bit represents the presence or absence of at least one photoelectron in a photodetector. A series of bit planes is generated through high-speed readout, and a kernel or “cubicle” of bits (x,y,t) is used to create a single output image pixel. The size of the cubicle can be adjusted post-acquisition to optimize image quality. The specialized sub-diffraction-limit photodetectors in the QIS are referred to as “jots” and a QIS may have a gigajot or more, read out at 1000 fps, for a data rate exceeding 1Tb/s. Basically, we are trying to count photons as they arrive at the sensor. This paper reviews the Quanta Image Sensor (QIS) concept and its imaging characteristics. Recent progress towards realizing the QIS for commercial and scientific purposes is discussed. The QIS represents a possible major paradigm shift in image capture.

A high spatial resolution iterative reconstruction algorithm is proposed for a half cone beam CT (HCBCT) geometry. The proposed algorithm improves spatial resolution by explicitly accounting for image blurriness caused by different factors, such as extended X-ray source and detector response. The blurring kernel is estimated using the MTF slice of the Catphan phantom. The proposed algorithm is specifically optimized for the new Leksell Gamma Knife Icon (Elekta AB, Stockholm, Sweden) which incorporates the HCBCT geometry to accommodate the existing treatment couch while covering down to the base-of-skull in the reconstructed field-of-view. Image reconstruction involves a Fourier-based scaling simultaneous algebraic reconstruction technique (SART) coupled with total variation (TV) minimization and non-local mean denoising, solved using a split Bregman separation technique that splits the reconstruction problem into a gradient based updating step and a TV-based deconvolution algorithm. This formulation preserves edges and reduces the staircase effect caused by regular TV-penalized iterative algorithms. Our experiments indicate that our proposed method outperforms the conventional filtered back projection and TV penalized SART methods in terms of line pair resolution and retains the favorable properties of the standard TV-penalized reconstruction.

Echocardiography provides valuable information to diagnose heart dysfunction. A typical exam records several minutes of real-time cardiac images. To enable complete analysis of 3D cardiac strains, 4-D (3-D+t) echocardiography is used. This results in a huge dataset and requires effective automated analysis. Ultrasound speckle tracking is an effective method for tissue motion analysis. It involves correlation of a 3D kernel (block) around a voxel with kernels in later frames. The search region is usually confined to a local neighborhood, due to biomechanical and computational constraints. For high strains and moderate frame-rates, however, this search region will remain large, leading to a considerable computational burden. Moreover, speckle decorrelation (due to high strains) leads to errors in tracking. To solve this, spatial motion coherency between adjacent voxels should be imposed, e.g., by averaging their correlation functions. 1 This requires storing correlation functions for neighboring voxels, thus increasing memory demands. In this work, we propose an efficient search using PatchMatch, 2 a powerful method to find correspondences between images. Here we adopt PatchMatch for 3D volumes and radio-frequency signals. As opposed to an exact search, PatchMatch performs random sampling of the search region and propagates successive matches among neighboring voxels. We show that: 1) Inherently smooth offset propagation in PatchMatch contributes to spatial motion coherence without any additional processing or memory demand. 2) For typical scenarios, PatchMatch is at least 20 times faster than the exact search, while maintaining comparable tracking accuracy.

Imaging in a bright field or phase contrast microscope is partially coherent. We have found that the image can be conveniently considered and modeled in terms of the Wigner distribution function (WDF) of the object transmission. The WDF of the object has a simple physical interpretation for the case of a slowly varying object. Basically, the image intensity is the spatial marginal of the spatial convolution of the object WDF with the phase space imager kernel (PSIkernel), a rotated version of the transmission cross-coefficient. The PSI-kernel can be regarded as a partially-coherent generalization of the point spread function. This approach can be extended to consider the partial coherence of the image itself. In particular, we can consider the mutual intensity, WDF or ambiguity function of the image. It is important to note that the spatial convolution of the object WDF with the PSI-kernel is not a WDF, and not the WDF of the image. The phase space representations of the image have relevance to phase reconstruction methods such as phase space tomography, or the transport of intensity equation approach, and to the three-dimensional image properties.

In this paper we describe a Hardware Accelerator (HWA) for fast recursive approximation of separable convolution with exponential function. This filter can be used in many Image Processing (IP) applications, e.g. depth-dependent image blur, image enhancement and disparity estimation. We have adopted this filter RTL implementation to provide maximum throughput in constrains of required memory bandwidth and hardware resources to provide a power-efficient VLSI implementation.

Action classification using a Bag of Words (BoW) representation has shown computational simplicity and good performance, but the increasing number of categories, including actions with high confusion, and the addition of significant contextual information has led most authors to focus their efforts on the combination of image descriptors. In this approach we code the action videos using a BoW representation with diverse image descriptors and introduce them to the optimal SVM kernel as a linear combination of learning weighted single kernels. Experiments have been carried out on the action database HMDB and the upturn achieved with our approach is much better than the state of the art, reaching an improvement of 14.63% of accuracy.

In this paper, we investigate an application that integrates holistic appearance based method and feature based method for face recognition. The automatic face recognition system makes use of multiscale Kernel PCA (Principal Component Analysis) characterized approximated face images and reduced the number of invariant SIFT (Scale Invariant Feature Transform) keypoints extracted from face projected feature space. To achieve higher variance in the inter-class face images, we compute principal components in higher-dimensional feature space to project a face image onto some approximated kernel eigenfaces. As long as feature spaces retain their distinctive characteristics, reduced number of SIFT points are detected for a number of principal components and keypoints are then fused using user-dependent weighting scheme and form a feature vector. The proposed method is tested on ORL face database, and the efficacy of the system is proved by the test results computed using the proposed algorithm.

In this paper, we present a hierarchical kernel associative memory (H-KAM) based computational model with Finite Ridgelet Transform (FRIT) representation for multispectral palmprint recognition. To characterize a multispectral palmprint image, the Finite Ridgelet Transform is used to achieve a very compact and distinctive representation of linear singularities while it also captures the singularities along lines and edges. The proposed system makes use of Finite Ridgelet Transform to represent multispectral palmprint image and it is then modeled by Kernel Associative Memories. Finally, the recognition scheme is thoroughly tested with a benchmarking multispectral palmprint database CASIA. For recognition purpose a Bayesian classifier is used. The experimental results exhibit robustness of the proposed system under different wavelengths of palm image.

With the same decorrelation and coding gain capabilities as the Karhunen-Loeve transform, the prediction-based lower triangular transform (PLT) can apply its perfect reconstruction property for lossless compression of ultraspectral sounder data. As the compression process requires computation of the covariance matrix, the LDU decomposition, and the transform kernel and coefficient matrices, it will be beneficial to introduce parallel processing technology in the PLT implementation. In this work, the recent Intel Many Integrated Core (MIC) architecture will be used which can exploit up to 60 cores with 4 hardware threads per core. Both threading and vectorization techniques will be explored for performance improvement. In our experiment, the total processing time of an AIRS granule can have a speedup of ~4.6x. With the offload mode, the MIC architecture provides a convenient and efficient tool for parallelization of the PLT compression scheme.

This paper presents a novel compressed histogram attribute profile (CHAP) for classification of very high resolution remote sensing images. The CHAP characterizes the marginal local distribution of attribute filter responses to model the texture information of each sample with a small number of image features. This is achieved based on a three steps algorithm. The first step is devoted to provide a complete characterization of spatial properties of objects in a scene. To this end, the attribute profile (AP) is initially built by the sequential application of attribute filters to the considered image. Then, to capture complete spatial characteristics of the structures in the scene a local histogram is calculated for each sample of each image in the AP. The local histograms of the same pixel location can contain redundant information since: i) adjacent histogram bins can provide similar information; and ii) the attributes obtained with similar attribute filter threshold values lead to redundant features. In the second step, to point out the redundancies the local histograms of the same pixel locations in the AP are organized into a 2D matrix representation, where columns are associated to the local histograms and rows represents a specific bin in all histograms of the considered sequence of filtered attributes in the profile. This representation results in the characterization of the texture information of each sample through a 2D texture descriptor. In the final step, a novel compression approach based on a uniform 2D quantization strategy is applied to remove the redundancy of the 2D texture descriptors. Finally the CHAP is classified by a Support Vector Machine classifier with histogram intersection kernel that is very effective for high dimensional histogram-based feature representations. Experimental results confirm the effectiveness of the proposed CHAP in terms of computational complexity, storage requirements and classification accuracy when compared to the other AP-based methods.

Image blind deconvolution is a more practical inverse problem in modern imaging sciences including consumer photography, astronomical imaging, medical imaging, and microscopy imaging. Among all of the latest blind deconvolution algorithms, the total variation based method provides privilege for large blur kernel. However, the computation cost is heavy and it does not handle the estimated kernel error properly. Otherwise, the using of the whole image to estimate the blur kernel is inaccurate because of that the insufficient edges information will hazard the accuracy of estimation. Here, we proposed a robust multi-frame images blind deconvolution algorithm to handle this complicated imaging model and applying it to the engineering community. In our proposed method, we induced the patch and kernel selection scheme to selecting the effective patch to estimate the kernel without using the whole image; then an total variation based kernel estimation algorithm was proposed to estimate the kernel; after the estimation of blur kernels, a new kernel refinement scheme was applied to refine the pre-estimated multi-frame estimated kernels; finally, a robust non-blind deconvolution method was implemented to recover the final latent sharp image with the refined blur kernel. Objective experiments on both synthesized and real images evaluate the efficiency and robustness of our algorithm and illustrate that this approach not only have rapid convergence but also can effectively recover high quality latent image from multi-blurry images.

The method for estimating the space-varying kernel for rotation motion which can’t be solved by a single kernel is proposed. After layering the image by multiplex difference of Gaussian model, the method estimate the rotary direction and the movement for each block in Fourier domain. An improved optimization for both direction and scale of different parts around the rotation center estimated with the same radius, through a constraint of United Least square Filter is taken in our algorithm to structure the blur path accurately. Aiming at the different position of the rotary region, combining the blur distribution estimated with an operator which created related to the spatial location, character and degree of rotation motion, here build a model to estimate a space-varying kernel for the rotation motion which is replaced by such pixels on the motion-blur-path with uniform and separate influence during the exposure in the original algorithm, it also could be used for space-variant de-blurring. Experimental results for synthetic and real images demonstrate the effectiveness of this algorithm.

Optical-coherence tomography (OCT) is a promising non-invasive, high-resolution imaging modality which can be used for cancer diagnosis and its therapeutic assessment. However, speckle noise makes detection of cancer boundaries and image segmentation problematic and unreliable. Therefore, to improve the image analysis for a precise cancer border detection, the performance of different image processing algorithms such as mean, median, hybrid median filter and rotational kernel transformation (RKT) for this task is investigated. This is done on OCT images acquired from an ex-vivo human cancerous mucosa and in vitro by using cultivated tumour applied on organotypical hippocampal slice cultures. The preliminary results confirm that the border between the healthy and the cancer lesions can be identified precisely. The obtained results are verified with fluorescence microscopy. This research can improve cancer diagnosis and the detection of borders between healthy and cancerous tissue. Thus, it could also reduce the number of biopsies required during screening endoscopy by providing better guidance to the physician.

The similarity measurement of the bilateral filtering can’t indicate the difference between pixels accurately in the dense texture region. It causes the smoothing effect seriously, thereby reducing the edge preserving properties of the bilateral filtering. This paper presents a new weighted function of the bilateral filtering. It involves an additional range kernel using the relative difference between pixels. The range kernel operates differently by acting on the pixel gray intensities or colors. And it uses the reciprocal kernel for the approximation of the standard Gaussian weight value. Experimental results suggest that it significantly preserves more details than the classical bilateral filtering in edge or dense texture regions.

Surveillance cameras have become ubiquitous in society, used to monitor areas such as residential blocks, city streets, university campuses, industrial sites, and government installations. Surveillance footage, especially of public areas, is frequently streamed online in real time, providing a wealth of data for computer vision research. The focus of this work is on detection of anomalous patterns in surveillance video data recorded over a period of months to years. We propose an anomaly detection technique based on support vector data description (SVDD) to detect anomalous patterns in video footage of a university campus scene recorded over a period of months. SVDD is a kernel-based anomaly detection technique which models the normalcy data in a high dimensional feature space using an optimal enclosing hypersphere – samples that lie outside this boundary are detected as outliers or anomalies. Two types of anomaly detection are conducted in this work: track-level analysis to determine individual tracks that are anomalous, and day-level analysis using aggregate scene level feature maps to determine which days exhibit anomalous activity. Experimentation and evaluation is conducted using a scene from the Global Webcam Archive.

Data fusion is a powerful theory that often leads to signi“cant performance gain and/or improved robustness of a givensolution. In this article, we explore how fusion can be used to advance our previously established improved EvolutionaryCOnstructed (iECO) imagedescriptor framework. The goalof iECO isto learn adiverse set ofindividuals (variable lengthchromosome in a genetic algorithm). Each iECO individual encodes a unique composition of different low-level imagetransformations in the context of a high-level image descriptor. Herein, we investigate multiple kernel (MK) aggregationand MK learning (MKL) for feature-levelŽ fusion of iECO chromosomes. Speci“cally, we explore MKL group lasso(MKLGL) and we put forth a new way to directly assign kernel weights from a mesure de“ned on the kernel matrices. Theproposed work is presented in the context of buried explosive hazard detection (EHD) in forward looking (FL) imagery.Experiments are reported using receiver operating characteristic (ROC) curves on data from a U.S. Army test site tha tcontains multiple target and clutter types, burial depths and times of day. We demonstrate that MK support vector machine(MKSVM) classi“cation outperform single kernel SVM (SKSVM) classi“cation and our weight assignment proceduregeneralizes well and outperforms MKLGL for EHD in FLIR.Keywords: Explosive hazard detection, iECO, feature learning, multiple kernel, long-wave infrared

In a previous study, we developed a hybrid tumor detection method that used both computed tomography (CT) and positron emission tomography (PET) images. However, similar to existing computer-aided detection (CAD) schemes, it was difficult to detect low-contrast lesions that touch to the normal organs such as the chest wall or blood vessels in the lung. In the current study, we proposed a novel lung tumor detection method that uses active contour filters to detect the nodules deemed "difficult" in previous CAD schemes. The proposed scheme detects lung tumors using both CT and PET images. As for the detection in CT images, the massive region was first enhanced using an active contour filter (ACF), which is a type of contrast enhancement filter that has a deformable kernel shape. The kernel shape involves closed curves that are connected by several nodes that move iteratively in order to enclose the massive region. The final output of ACF is the difference between the maximum pixel value on the deformable kernel, and pixel value on the center of the filter kernel. Subsequently, the PET images were binarized to detect the regions of increased uptake. The results were integrated, followed by the false positive reduction using 21 characteristic features and three support vector machines. In the experiment, we evaluated the proposed method using 100 PET/CT images. More than half of nodules missed using previous methods were accurately detected. The results indicate that our method may be useful for the detection of lung tumors using PET/CT images.

Purpose: Conventional workflow in interventional imaging often ignores a wealth of prior information of the patient anatomy and the imaging task. This work introduces a task-driven imaging framework that utilizes such information to prospectively design acquisition and reconstruction techniques for cone-beam CT (CBCT) in a manner that maximizes task-based performance in subsequent imaging procedures. Methods: The framework is employed in jointly optimizing tube current modulation, orbital tilt, and reconstruction parameters in filtered back-projection reconstruction for interventional imaging. Theoretical predictors of noise and resolution relates acquisition and reconstruction parameters to task-based detectability. Given a patient-specific prior image and specification of the imaging task, an optimization algorithm prospectively identifies the combination of imaging parameters that maximizes task-based detectability. Initial investigations were performed for a variety of imaging tasks in an elliptical phantom and an anthropomorphic head phantom. Results: Optimization of tube current modulation and view-dependent reconstruction kernel was shown to have greatest benefits for a directional task (e.g., identification of device or tissue orientation). The task-driven approach yielded techniques in which the dose and sharp kernels were concentrated in views contributing the most to the signal power associated with the imaging task. For example, detectability of a line pair detection task was improved by at least three fold compared to conventional approaches. For radially symmetric tasks, the task-driven strategy yielded results similar to a minimum variance strategy in the absence of kernel modulation. Optimization of the orbital tilt successfully avoided highly attenuating structures that can confound the imaging task by introducing noise correlations masquerading at spatial frequencies of interest. Conclusions: This work demonstrated the potential of a task-driven imaging framework to improve image quality and reduce dose beyond that achievable with conventional imaging approaches.

Combined PET/MR imaging allows to incorporate the high-resolution anatomical information delivered by MRI into the PET reconstruction algorithm for improvement of PET accuracy beyond standard corrections. We used the working hypothesis that glucose uptake in adipose tissue is low. Thus, our aim was to shift 18 F-FDG PET signal into image regions with a low fat content. Dixon MR imaging can be used to generate fat-only images via the water/fat chemical shift difference. On the other hand, the Origin Ensemble (OE) algorithm, a novel Markov chain Monte Carlo method, allows to reconstruct PET data without the use of forward- and back projection operations. By adequate modifications to the Markov chain transition kernel, it is possible to include anatomical a priori knowledge into the OE algorithm. In this work, we used the OE algorithm to reconstruct PET data of a modified IEC/NEMA Body Phantom simulating body water/fat composition. Reconstruction was performed 1) natively, 2) informed with the Dixon MR fat image to down-weight 18 F-FDG signal in fatty tissue compartments in favor of adjacent regions, and 3) informed with the fat image to up-weight 18 F-FDG signal in fatty tissue compartments, for control purposes. Image intensity profiles confirmed the visibly improved contrast and reduced partial volume effect at water/fat interfaces. We observed a 17±2% increased SNR of hot lesions surrounded by fat, while image quality was almost completely retained in fat-free image regions. An additional in vivo experiment proved the applicability of the presented technique in practice, and again verified the beneficial impact of fat-constrained OE reconstruction on PET image quality.

Static C-Arm CT 3D FDK baseline reconstructions (3D-DSA) are unable to provide temporal information to radiologists. 4D-DSA provides a time series of 3D volumes implementing a constrained image, thresholded 3D-DSA, reconstruction utilizing temporal dynamics in the 2D projections. Volumetric limiting spatial resolution (VLSR) of 4DDSA is quantified and compared to a 3D-DSA reconstruction using the same 3D-DSA parameters. Investigated were the effects of varying over significant ranges the 4D-DSA parameters of 2D blurring kernel size applied to the projection and threshold applied to the 3D-DSA when generating the constraining image of a scanned phantom (SPH) and an electronic phantom (EPH). The SPH consisted of a 76 micron tungsten wire encased in a 47 mm O.D. plastic radially concentric thin walled support structure. An 8-second/248-frame/198 ° scan protocol acquired the raw projection data. VLSR was determined from averaged MTF curves generated from each 2D transverse slice of every (248) 4D temporal frame (3D). 4D results for SPH and EPH were compared to the 3D-DSA. Analysis of the 3D-DSA resulted in a VLSR of 2.28 and 1.69 lp/mm for the EPH and SPH respectively. Kernel (2D) sizes of either 10x10 or 20x20 pixels with a threshold of 10% of the 3D-DSA as a constraining image provided 4D-DSA VLSR nearest to the 3D-DSA. 4D-DSA algorithms yielded 2.21 and 1.67 lp/mm with a percent error of 3.1% and 1.2% for the EPH and SPH respectively as compared to the 3D-DSA. This research indicates 4D-DSA is capable of retaining the resolution of the 3D-DSA.

Scattered radiation remains one of the primary challenges for digital mammography, resulting in decreased image contrast and visualization of key features. While anti-scatter grids are commonly used to reduce scattered radiation in digital mammography, they are an incomplete solution that can add radiation dose, cost, and complexity. Instead, a software-based scatter correction method utilizing asymmetric scatter kernels is developed and evaluated in this work, which improves upon conventional symmetric kernels by adapting to local variations in object thickness and attenuation that result from the heterogeneous nature of breast tissue. This fast adaptive scatter kernel superposition (fASKS) method was applied to mammography by generating scatter kernels specific to the object size, x-ray energy, and system geometry of the projection data. The method was first validated with Monte Carlo simulation of a statistically-defined digital breast phantom, which was followed by initial validation on phantom studies conducted on a clinical mammography system. Results from the Monte Carlo simulation demonstrate excellent agreement between the estimated and true scatter signal, resulting in accurate scatter correction and recovery of 87% of the image contrast originally lost to scatter. Additionally, the asymmetric kernel provided more accurate scatter correction than the conventional symmetric kernel, especially at the edge of the breast. Results from the phantom studies on a clinical system further validate the ability of the asymmetric kernel correction method to accurately subtract the scatter signal and improve image quality. In conclusion, software-based scatter correction for mammography is a promising alternative to hardware-based approaches such as anti-scatter grids.

Face detection is an important aspect for biometrics, vide o surveillance and human comput er interaction. Due to the complexity of the detection algorithms any face detection system requires a huge amount of computational and memory resources. In this communication an accelerated implementation of MB LBP face detection algorithm targeting low frequency, low memory and low power embedded system is presented. The resulted implementation is time deterministic and uses a customizable AMBA IP hardware accelerator. The IP implements the kernel operations of the MB-LBP algorithm and can be used as universa l accelerator for MB LBP based applicati ons. The IP employs 8 parallel MB-LBP feature evaluators cores, uses a deterministic bandwidth, has a low area profile and the power consumption is ~95 mW on a Virtex5 XC5VLX50T. The resulted implementation accelera tion gain is between 5 to 8 times, while the hardware MB-LBP feature evaluation gain is between 69 and 139 times. Keywords: Face detection, AMBA, LBP, OpenCV, FPGA.

This paper presents a new 3D mesh smoothing algorithm which is based on evolutionary methods. This method is a new optimization technique with Genetic algorithm. The main approach is based on expanding the search space by generating new meshes as genetic individuals. Features and shrinkage of models are preserved that are the main problems of existing smoothing algorithms. So with this method, over-smoothing effects are reduced and undesirable noises are effectively removed. Keywords: smoothing, 3D Meshing, Laplacian, Genetic Algorithms. 1. INTRODUCTION With the fast development of 3d acqusition devices, 3d models are being used in many fields such as medicine, the media, art and entertainment. 3d scanners analyze the shape of real-world objects and coll ects point clouds for geometric samples of surfaces. These point clouds are usually represen ted with triangle meshes. For many reasons 3D model can contain noise and it must be smoothed before the further mesh processing. The mesh smoothing is an intensive subject in computer graphics, computer aided design (CAD) and other related fields [1]. One of the well known approach for smoothing is Laplacian algorithm. The basic idea for Laplacian algorithm is to set the location of a vertex. This is performed by calculating a new location as the average of the locations of vertices in the neighborhood. The process is continued until there is sufficient smoothing. Shrinking is the essential difficulty of Laplacian algorithm. Many studies are carri ed out in the past years based on Laplacian algorithm and solve the shrinking problem.[2-4] Laplacian flow is also called as Lowpass filtering[2]. Mean, median and bliteral filters are also applied to mesh smoothing problem and also adopted to image processing[5]. Many other approaches have been introduced, their main goal is to define a cost function related with mesh elements like; positions of vertices, edge lengths, triangle areas, angles. Algorithms have been developed to minimize this cost functions [6]. Mesh smoothing has also been dealt with physical analogy point of view [7]. Kernel-Based Laplacian[8] and local curvature point-set[9] techniques are other related works of smoothing. Optimization based techniques are also used to solve smoothing problem. A combination of constrained Laplacian smoothing together with an optimization-based smoothing algorithm are represented by Canann et. al [10] For reducing the shrinking effect of Laplacian flow, an optimization based function is introduced by Hamam et.al [11] In this paper a new optimization method is presented for surface smoothing. The proposed method is based on Genetic algorithms and Evolutionary process, which is the specification of reposition vertices of models as a new chromosome. In optimization process, new individuals are generated for solution space by using crossover and mutation operators. The mutation process is based on a random vertice displacement of randomly choosen individual. The smoothing operation is based on evolution of new individuals.

In this investigation we identify relationships between human and automated face recognition systems with respect to compression. Further, we identify the most influential scene parameters on the performance of each recognition system. The work includes testing of the systems with compressed Closed-Circuit Television (CCTV) footage, consisting of quantified scene (footage) parameters. Parameters describe the content of scenes concerning camera to subject distance, facial angle, scene brightness, and spatio-temporal busyness. These parameters have been previously shown to affect the human visibility of useful facial information, but not much work has been carried out to assess the influence they have on automated recognition systems. In this investigation, the methodology previously employed in the human investigation is adopted, to assess performance of three different automated systems: Principal Component Analysis, Linear Discriminant Analysis, and Kernel Fisher Analysis. Results show that the automated systems are more tolerant to compression than humans. In automated systems, mixed brightness scenes were the most affected and low brightness scenes were the least affected by compression. In contrast for humans, low brightness scenes were the most affected and medium brightness scenes the least affected. Findings have the potential to broaden the methods used for testing imaging systems for security applications.

A first diagnosis of colorectal cancer is performed by examination of polyp shape and appearance during an endoscopy routine procedure. However, the video-endoscopy is highly noisy because exacerbated physiological conditions like increased motility or secretion may limit the visual analysis of lesions. In this work a 3D reconstruction of the digestive tract is proposed, facilitating the polyp shape evaluation by highlighting its surface geometry and allowing an analysis from different perspectives. The method starts by a spatio-temporal map, constructed to group the different regions of the tract by their similar dynamic patterns during the sequence. Then, such map was convolved with a second derivative of a Gaussian kernel that emulates the camera distortion and allows to highlight the polyp surface. The position initialization in each frame of the kernel was computed from expert manual delineation and propagated along the sequence based on. Results show reliable reconstructions, with a salient 3D polyp structure that can then be better observed.

Restoring blurred images is challenging because both the blur kernel and the sharp image are unknown, which makes this problem severely under constrained. Recently many single image blind deconvolution methods have been proposed, but these state-of-the-art single image deblurring techniques are still sensitive to image noise, and can degrade their performance rapidly especially when the noise level of the input blurred images increases. In this work, we estimate the blur kernel accurately by applying a series of directional low-pass filters in different orientations to the input blurred image, and effectively constructing the Radon transform of the blur kernel from each filtered image. Finally, we use a robust non-blind deconvolution method with outlier handling, which can effectively reduce ringing artifacts, to generate the final results. Our experimental results on both synthetic and real-world examples show that our method achieves comparable quality to existing approaches on blurry noisy-free images, and higher quality outputs than previous approaches on blurry and noisy images.

A temporal-spatial filtering algorithm based on kernel density estimation structure is presented for background suppression in this paper. The algorithm can be divided into spatial filtering and temporal filtering. Smoothing process is applied to the background of an infrared image sequence by using the kernel density estimation algorithm in spatial filtering. The probability density of the image gray values after spatial filtering is calculated with the kernel density estimation algorithm in temporal filtering. The background residual and blind pixels are picked out based on their gray values, and are further filtered. The algorithm is validated with a real infrared image sequence. The image sequence is processed by using Fuller kernel filter, Uniform kernel filter and high-pass filter. Quantitatively analysis shows that the temporal-spatial filtering algorithm based on the nonparametric method is a satisfactory way to suppress background clutter in infrared images. The SNR is significantly improved as well.

Accurate Point Spread Function (PSF) estimation of coded aperture cameras is a key to deblur defocus images. There are mainly two kinds of approaches to estimate PSF: blind-deconvolution-based methods, and measurement-based methods with point light sources. Both these two kinds of methods cannot provide accurate and convenient PSFs due to the limit of blind deconvolution or imperfection of point light sources. Inaccurate PSF estimation introduces pseudo-ripple and ringing artifacts which influence the effects of image deconvolution. In addition, there are many inconvenient situation for the PSF estimation. This paper proposes a novel method of PSF estimation for coded aperture cameras. It is observed and verified that the spatially-varying point spread functions are well modeled by the convolution of the aperture pattern and Gaussian blurring with appropriate scales and bandwidths. We use the coded aperture camera to capture a point light source to get a rough estimate of the PSF. Then, the PSF estimation method is formulated as the optimization of scale and bandwidth of Gaussian blurring kernel to fit the coded pattern with the observed PSF. We also investigate the PSF estimation at arbitrary distance with a few observed PSF kernels, which allows us to fully characterize the response of coded imaging systems with limited measurements. Experimental results show that our method is able to accurately estimate PSF kernels, which significantly make the deblurring performance convenient.

Fluctuations in the images of scenes viewed over large distances are the most obvious manifestation of the turbulence effects on the imaging of the incoherent objects. While the average or long-exposure imaging is arguably the most well studied topic of the optical propagation in turbulence, and substantial progress was also made in understanding the average short-exposure imaging, the image scintillations for complex extended scenes are not well understood. We discuss some available results of the image scintillation theory and report on some recent progress. We introduce the concept of the scintillation imaging, when unlike the conventional turbulence imaging techniques the variance of the series of images of the scene is calculated and used to gain information either about the object or about the turbulence on the propagation path. The third constraint in the turbulent PSF [1] plays a critical role in the scintillation imaging making scintillation images insensitive to the constant background and emphasizing the areas with higher local contrast. The bilinear structure of the Object-to-Variance (O2V) maps makes it impossible to use the analogues of the PSF or MTF for scintillation images and precludes development of the general theory of scintillation imaging. We discuss the fundamental properties of the O2V kernel and discuss four examples of scintillation images of simple objects. We present the measurement data where colored scintillation images of the edge were obtained. The variance distributions are normalized using the traditional long-exposure images to remove dependence on the object brightness. In this case scintillations are concentrated near the edge and carry information about the turbulence on the imaging path. The amplitude and width of these variance distributions are sensitive to the turbulence level and can be used as passive scintillometer without the need to deploy the laser source and receiver at both ends of the propagation path. Variance images of the object with sinusoidal brightness distribution consists of the uniform background and doublefrequency sinusoidal oscillations. It has the features consistent with turbulent super-resolution originally described in [2]. Namely, for unresolved object oscillating components disappears while the background persevere.