Research On Correlation Noise Model And WZ Frame Reconstruction In Distributed Video Coding

With the rapid development of information technology, a new up-link oriented videoapplication such as wireless video surveillance, multimedia sensor network, mobile cameraphone, and so on, is popping up. Encoding devices are relatively simple equipment that islimited by the power supply, power, computing resources in these application fields.Traditional video coding standard, such as MPEG series and H.26x series, is no longersuitable for these new applications, so it is necessary to find and research new video codingstandard. Distributed video coding (DVC) provides a good solution for these new videoapplications, thus it becomes a domestic and foreign research hotspot. Correlation noisemodel (CNM) and WZ frame reconstruction algorithm in DVC are key factors influencing thecoding performance of the system. This dissertation mainly analyzes the technicalcharacteristics of CNM and WZ frame reconstruction algorithm, deeply studies on exitingproblems, and puts forward the corresponding solutions. The main research work is asfollows:1. The lossless Slepian-Wolf theorem and the lossy Wyner-Ziv theorem of distributedsource coding (DSC) are described. Several typical DVC frameworks such as pixel-domainWyner-Ziv system, transform-domain Wyner-Ziv system, DISCOVER, PRISM, and so on, areanalyzed. The latest research progresses of CNM and WZ frame reconstruction algorithm arereviewed.2.To study the Laplacian distribution and generalized Gaussian distribution (GGD), weanalyse the characteristics of these two kinds of CNM and expound their parameter estimationmethods. Moreover, we analyse the characteristics of generalized Gamma (GGA) distribution,study the performance of GGA distribution fitting correlation noise, present the parameterestimation algorithm of GGA distribution. Experimental results show that, GGA distributionovermaches Laplacian distribution and GGD in terms of fitting correlation noise.3. Having studied the performance of GGA distribution fitting correlation noise, theDVC CNM using GGA distribution is proposed. The DVC systems of CNM using GGAdistribution, Laplacian distribution, GGD are respectively implemented. Experimental results prove that, as compared with using Laplacian distribution or GGD, the CNM using GGAdistribution effectively improves the rate distortion performance and significantly savesdecoding time of DVC system.4. Minimum mean square error (MMSE) reconstruction algorithm is a kind ofreconstruction algorithm based on Laplacian distribution. According to experimental results,the performance of GGD fitting correlation noise is better than that of Laplacian distribution,thus WZ frame reconstruction algorithm based on GGD in DVC system is proposed. Theclosed-form expressions of reconstruction are derived for CNM using GGD with shapeparameter fixed to0.5. Experimental results indicate that, as compared with straightreconstruction algorithm and MMSE reconstruction algorithm, the proposed reconstructionalgorithm effectively improves the image quality of the decoded WZ frame in DVC system.