Research On Visual Representation Enhancement Of Digital Image And Video Contents

In the process of image acquisition, storage and display, there are many factors that can cause image visual quality deterioration, and the visual quality of digital content tends to degradation. For example, video quality will be compromised by jitter or shake while videos captured with a handheld video camera, and this often partly has color cast. The visual quality of the image content is incontrovertibly reduced when performing color-to-gray process, the reason is the discarded color information. An in-depth study is carried out in this dissertation to reduce the visual quality of digital content degradation with novel model and effective solution proposed.Camera shake is one of the most common reasons for image blur. This dissertation introduces spatial-variant upper and lower bound constraints to regularize Total Vari-ation blind deconvolution. The local upper and lower bound constraints are computed based on the local structure of the observed image. In order to suppress the ringing ar-tifacts in the common deblurred images, this paper introduces a learning based method to model the distribution function of the ringing patches. The Gaussian Mixture Model is used to fit the distribution. Then the learned probability model functions as a prior for improving the deblurring quality. Experimental results proves the proposed algo-rithm to be effective to image deblurring with high quality and suppress the ringing artifact in the deblurred image.Color is critical in visual expressions. In this dissertation, leveraging cross-channel information for color constancy algorithm is proposed. Color cast is generally occurred when taking photos due to the camera imaging sensor response characteristics of light. In order to restore original color of the objects in the scene, color information of the im-age is corrected by exploiting cross-channel of the local and global spectral character- istics of the visual image data. Experimental results indicates algorithm effectiveness on varies of real images.Moreover, jitter or shake will occur in digital videos when it is produced with hand-held equipment, and wobble will occur if cameraman is on foot. Digital video stabilization improved visual quality through removing camera movements. In this dis-sertation, a model is proposed to remove camera jitter with a three-layer framework for video frame matching, video shake removal and video completion, respectively. The proposed algorithm can yield well shake removal and obtain satisfactory visual quality of video content with high calculation speed and low storage requirement, which is validated in actual test.Finally, decolorization, or color-to-gray, is widely applied in single-channel im-age processing. Decolorization is an essentially dimensionality reduction problem and hence is difficult. Transforming a color image to a grayscale one inevitably suffers from information loss. This dissertation proposed a novel method for converting the color images to the grayscale ones, aiming at preserving the color contrast. The intensity values are globally determined. An improved localized modification for human visual system targets on the accuracy of color difference measurement. Chrominance edges are enhanced by guided-filter, and a bimodal objective function is employed to relax the stringent order constraint for color mapping and allows for selecting suitable gray values automatically for preserving the original color contrast. Both the quantitative and qualitative evaluations demonstrate the effectiveness of our proposed approach.