An Algorithm of Wavelet Image Fusion Based on Visual Saliency

Considering the characteristics of wavelet transform and human visual system saliency, a novel image fusion algorithm was proposed. Based on multi-scale wavelet decomposition, visual saliency model had been introduced into calculating the intensity, direction and texture global integrated saliency of each pixel at each level frequency component. And it adaptively acquired the weighted fusing coefficients. Experiments show that this algorithm can achieve more clear minutia. Comparing with regional feature-based wavelet fusion algorithm, it improves fusion performance of human vision and objective evaluation. It is also benefit to detection and recognition.     

A New Image Segmentation for Virtual Background Reconstruction

This paper presents a new method of background segmentation for virtual background reconstruction in video-based virtual reality application. To aim at discounting the effect of non-uniform lightness on the diffuse and specular reflection background respectively, a background-colour measurement for diffuse as well as specular reflection is proposed based on Phong illumination model. This colour estimation has two steps： first, based on the colour invariance of diffuse reflection, the diffuse component of the model was estimated; second, the specular component was derived by using estimated diffuse factor and differerial laws. The experimental results of background segmentation show that the presented algorithm is considered robust to non-uniform lightness for diffuse reflection as well as the specular reflection and is also applicable for real-time vldeo-based virtual reality system.     

Improving Integrality Objects Based on of Detected Moving Image Matting

The integrality of moving objects is the basis for video-based object tracking and action analysis. But it often becomes unreliable as a result of shadow elimination when the object has similar properties with real shadow. The proposed methods manage to improve the integrality of detected moving objects as much as possible, by image matting operated on candidate shadow regions. Existing approaches for image matting require manual labeling of foreground and background. Considering the moving feature points in shadow may cause classification errors, we propose automatic scribbling methods based on Scale invariant feature transform （SIFT） and Speeded-up robust feature （SURF） respectively. Experiments demon- strate that our methods eliminate real shadow effectively and improve object segmentation in the case of object parts and shadows presenting similar characteristics.     

Weighted Discriminative Sparse Coding for Image Classification

In the discriminative sparse coding, the reconstruction residual over each class-specific sub- dictionary can provide great discriminative and label in- formation. In this paper, we propose a weighted discrim- inative sparse coding method by using the residual as the weight. For a test sample, we first compute its sparse code over each learnt sub-dictionary, and then use the recon- struction residual over each sub-dictionary to weight the corresponding sub-dictionary,thereby forming a weighted sample-specific structure dictionary, over which we com- pute a new sparse code for the test sample. This code car- ries more discriminative information about interclass dif- ference. Our method yields a unique structure dictionary for each test sample, so that samples with the same class labels have more similar distributions of dictionary atom contributions. Experimental results demonstrate that the proposed method outperforms some state-of-the-art meth- ods under the same learning conditions.     

Noise Reduction for InSAR Phase Images Using BM3D

This paper extends the application of Block-matching （BM） and 3D transform-domain collabo- rative filtering （BM3D） to the noise reduction in Interfer- ometric synthetic aperture radar （InSAR） phase imagery, and proposes a denoising algorithm which can effectively remove noise and preserve fringes. Since the noise level estimated by a median estimator is not always optimal for proposed algorithm in wavelet domain, a method of calcu- lating optimal noise standard deviation is also developed. The proposed algorithm is efficient and robust. Experi- mental results show that the visual quality and evaluation indexes of proposed algorithm outperform other filters for both simulated and real InSAR images.     

Digital Image Forensics Using Orthogonal 1-D Objects

The imaging features are shared by all images shot by cameras, the artificial images may break these features. We introduce the image formation model to image forensics. Orthogonal 1-D objects with known scale on either direction are used to estimate the principal point parameter, and the inconsistency of the principal point indicates a forgery. Experiments show that our method can detect cropping and copy-paste efficiently, and it is robust to smoothing, noise and JPEG compression.     

An Improved Combined Multiscale Method for Image Denoising

This paper proposes an improved combined multiscale method for image denoising. It uses the adaptive Bayesian shrinkage on the Undecimated wavelet transform （UWT） coefficients to denoise homogeneous areas, and the hard thresholding on the Fast discrete curvelet transform （FDCT） coefficients to denoise areas with edges. At the same time, it adopted a simple method to fuse two filtered images by UWT and FDCT separately. The experimental results indicate that this method has better performance with easier implementation and lower computational complexity than the existing combined methods.     

Automatic Image Annotation Using Semi-Supervised Multi-Instance Multi-Label Learning Algorithm

Automatic image annotation has emerged as an important research topic. From the perspective of machine learning, the annotation task fits both multiinstance and multi-label learning framework due to the fact that an image is composed of multiple regions, and is associated with multiple keywords as well. In this paper, we propose a novel Semi-supervised multi-instance multi-label （SSMIML） learning framework, which aims at taking full advantage of both labeled and unlabeled data to address the annotation problem. Specifically, a reinforced diverse density algorithm is applied firstly to select the Instance prototypes （IPs） with respect to a given keyword from both positive and unlabeled bags. Then, the selected IPs are modeled using the Gaussian mixture model （GMM） in order to reflect the semantic class density distribution. Furthermore, based on the class distribution for a keyword, both positive and unlabeled bags are redefined using a novel feature mapping strategy. Thus, each bag can be represented by one fixed-length feature vector so that the manifold-ranking algorithm can be used subsequently to propagate the corresponding label from positive bags to unlabeled bags directly. Experiments on the Corel data set show that the proposed method outperforms most existing image annotation algorithms.     

PCNN Model Analysis and Its Automatic Parameters Determination in Image Segmentation and Edge Detection

The Pulse coupled neural network （PCNN） has been widely used in digital image processing, but the automatic parameters determination is still a difficult as- pect, which becomes the focus of PENN research. In this paper, by the classical solution to difference equations and the time-domain analysis of PCNN model, we provide the expressions of the firing time and the firing period of neu- rons, and reveal the “mathematics firing” phenomenon of PCNN. Based on this, we propose a new method of auto- matic parameters determination based on both eliminating the “mathematics firing” and getting the highest efficiency of PCNN. We also present an edge detection model on the basis of image segmentation of PCNN and a method to determine automatically the parameters of the model. Ex- perimental results prove the validity and efficiency of our proposed algorithm for the segmentation and the edge de- tection of the test images.     

A Novel Coupled Bi-directional Diffused Image Inpainting Model with Morphological Invariant Property

In this paper, a novel coupled inpainting model which bi-directionally diffuses image information is proposed. It is morphological invariant which restores the target region based on geometric property. The image information is diffused along both the direction normal to edges and along isophotes. Total variation （TV） model is used to diffuse along the direction normal to edges which reconnects the broken lines in the target region directly. The along isophotes diffused part is the inviscid Helmholtz vorticity equation in fluid dynamics. The Helmholtz equation diffuses a smooth measure of image along isophotes, and it is morphological invariant. The novel model is comprised by 2nd order Partial differential equations （PDEs）, so the numerical scheme of it is simple and the processing time is limited. Experimental results demonstrate that the novel model smoothly restores the target region by diffusing along two orthogonal directions and preserves the linear structure which leads to its better performance than conventional inpainting models.     

Image Registration Using a Complete Polar Transform

Image registration is widely used in im- age processing. Researchers have introduced image reg- istration techniques based on the log-polar transform for its rotation and scale invariant properties. It suffers from nonuniform sampling which makes the registration results susceptible to interference. To address the problems of traditional log-polar transform, a Complete polar trans- form （CPT） method is proposed, which samples the image evenly to preserve the whole information of original image. An innovative projection transform is applied after CPT to obtain the rotation invariant property. We pre-matched feature points using the Scale invariant feature transform （SIFT）algorithm and re-matched them based onCPT to improve matching speed and accuracy. Experimental re- sults show that the proposed method is accurate and ro- bust to noise and alteration.     

基于正则化方向失真的压缩图像质量评价

图像质量评价是图像和视频处理系统中的基本算法。该文在分析人眼视觉系统对图像感知特点的基础上,结合二维图像的几何本征特征研究,提出了一种新型的基于图像正则化方向失真的完全参考质量评价算法。该方法计算简单,物理意义明确。仿真试验结果表明,该方法符合人眼视觉系统特性,所得到的图像质量客观评价结果与主观评价结果之间具有更好的一致性。     

Hybrid Weighted l1-Total Variation Constrained Reconstruction for MR Image

Compressed sensing based Magnetic resonance （MR） image reconstruction can be done by minimizing the sum of least square data fitting item, the Total variation （TV） and ll norm regularizations. In this paper, inspired by the conventional constrained reconstruction model, we propose a hybrid weighted l1-TV minimization method to reconstruct MR image. We introduce the iterative mechanism to update the weights for l1 and TV norms adaptively. The weights vary at each element of the image matrix according to the presented weights selection strategy. Experiments on Shepp-Logan phantom and practical MR. images demonstrate the proposed method can preserve image details and obtain improved reconstruction quality compared to the traditional CS-MRI method and other weighted methods.     

A Novel Adaptive Algorithm for Real-Time Image Enhancement

Aiming at the limitation of the traditional algorithm of unsharp masking for image enhancement, we deduce the function to get the coefficient for controlling image enhancement based on the unsharp masking and Gaussian operator. Then we propose a novel adaptive algorithm for real-time image enhancement, and analyze the way for realizing the algorithm by hardware. At last contradistinctive experimental results are given of the algorithm and other algorithms for image enhancement. The results show that we can enhance detail regions of image to the greatest extent, avoid excessive overshoot in edge regions, and minimize the degree of enhancement in uniform regions. A pleasant result is achieved by using the proposed algorithm. Furthermore, it can be realized simply by hardware. Accordingly, we can enhance image effectively to satisfy real-time demand.     

基于小波的方向自适应图像插值

图像插值是图像处理的一项重要技术,经典的插值算法会产生细节模糊和边缘锯齿现象。该文提出一种基于小波的方向自适应图像插值方法,将小波变换思想和局部方向自适应插值方法结合。为了获得清晰的细节部分,对图像实施改进的方向自适应双线性插值;结合小波方法,提高插值图像的高频细节信息,并进行相关后处理,增强视觉效果。实验结果表明,该文方法插值后的图像边缘清晰光滑,有效抑制了边缘模糊和锯齿现象,相比较传统方法,插值图像的客观质量和视觉效果都得到明显增强,更加适合人眼视觉系统。     

CNN Based No-Reference HDR Image Quality Assessment

Motivated by the problems of non-universality and over-reliance on the original reference image in High dynamic range (HDR) Image quality assessment (IQA), a convolutional neural network-based algorithm for no-reference HDR image quality assessment is proposed. The Salience detection by self-resemblance (SDSR) algorithm which extracts the salient regions of the HDR image, is used to simulate the human visual attention mechanism. Then a visual quality perception network for training quality prediction models is designed according to the visual characteristics of luminance and contrast sensitivity. And this network consists of an Error estimation network (Error-net), a Perceptual resistance network (PR-net) and a mixing function. The experimental results indicate that the method proposed has high consistency with subjective perception, and the value of assessment metrics Spearman rank-order correlation coefficient (SROCC), Pearson product-moment correlation coefficient (PLCC) and Root mean square error (RMSE) correspondingly reaches 0.941, 0.910 and 8.176 as well. It is comparable with classic full-reference HDR IQA methods.     

Region-based Anisotropic Diffusion with Soft Shock Filter for Adaptive Image Enhancement

A region-based anisotropic diffusion with soft shock filter is presented for image noise removal and edge sharpening. Two image processing steps are performed successively： smoothing and sharpening. An image is divided into different regions according to image features： edges, textures and details, and fiat areas. For edges, a shocktype backward diffusion is performed in the gradient direction to the isophote line （edge）, incorporating a forward diffusion in the isophote line direction; while for textures and details, a soft backward diffusion is done to enhance image features preserving a natural transition. Experiments on real images show that this method produces better visual results of the enhanced images than some relevant equations.     

Ship Detection and Characterization with Generalized Optimization of Polarimetric Contrast Enhancement

A Generalized optimization algorithm of polarimetric contrast enhancement （GOPCE） is proposed to find optimal polarimetric states and coefficients to max-imize the received power ratio of ship to ocean clutter, which facilitates ship detection. The method is validated with a real SIR-C/X image. The results indicate that, rather than optimal polarimetric contrast enhancement, GOPCE apparently improves the contrast of ship to sea by introducing the polarimetric feature vector.     

图像融合中的图像配准方法研究

图像配准是对取自不同时间、不同传感器或不同视角的同一场景的两幅图像或者多幅图像匹配的过程。快速准确地实现配准是图像融合的前提,为此提出了基本相似性测度算法的模板匹配方法。对像素相关性小的红外和可见光图像,提出了基于sobel 快速边缘检测的模板匹配方法。