Research On Distributed Video Coding Based On WMSN

With the development of wireless multimedia sensor networks, there are moreand more Video applications. Because of the limited terminal node processingcapacity and network resources, low complexity encoding is demanded in theseapplications. Traditional video coding standards, encoder uses motion estimation andmotion compensation technology, the encoder is5to10times more complex thanthe decoder. However, the distributed video coding is based on independentencoding and joint deeoding and moves the implementation of motion estimationand motion offset to the decoder side, so as to simplify the eneoder. The distributedvideo coding is very applicable to solve the video coding issues on the wrirelessmultimedia sensor networks.This thesis mainly makes a study of distributed video coding based on WMSN.The improved key frame algorithm and improved side information algorithms areproposed after studying the transform domain DVC. The main work andachievements are as follows:(1) This paper studies the encoding of key frames and propose a key framescoding mode selection method based on band allocation, coefficient tape obtainedfrom DCT transform of the key word are divided into low frequency and highfrequency, Wyner-Ziv codec is applied to the low frequency; however, anappropriate coding mode is selected for the high frequency, taking Wyner-Ziv codecand traditional intraframe codec, respectively.(2)This paper studies two different channel coding: Turbo and LDPC, andstudies the applications in Slepian-Wolf encoder.This paper simulates theirperformance based on the transform domain DVC, the stimulation is that theperformance of LDPC is better than Turbo.(3)After studing the common side information generation algorithm, this thesisproposes a successive refinement side information generation algorithm based onDCT distributed video coding, initial side information which is generated by motioncompensated interpolation is futher refined to find unaccurate motion vector. Inorder to generate accurate side infortion, the motion vector will be corrected.Simulation results show that compared with the transform domain DVC, theproposed algorithm effectively improves the quality of side information, especially for video sequences with violent movement. The improvements are2.4dB and1.1dBrespectively for Foreman and Carphone.(4)This paper simulates the improved system performance, compared with thetransform domain DVC, the proposed algorithm has a good performance for PSNR.The improvement is greater than the transform domain DVC when compared withH.264intra coding,especially for video sequences with violent movement. Theimprovements are1-2dB and0.8-1dB respectively for Foreman and Carphonecompared with the transform domain DVC.