Research On Distributed Video Coding Based On Compressive Sensing

At present, the requirement for wireless video communication has grown dramatically alongwith the rapid development of wireless communication techniques. However, the traditional videocoding techniques applied widely in the cable network are facing great challenges in the wirelesscommunication system: it is the computation complexity and energy consumption of encoder in thevideo terminal that are limited strictly, and channel instability cannot be guaranteed. Therefore, thefuture wireless video communication requires to take the new theories and techniques in account forrealizing the efficient and reliable video compression. Distributed Video Coding (DVC) is a validmethod in the wireless communication domain and has the ability to shift computation complexityfrom encoder to decoder, hence it has the advantages of low encoding complexity, high compres-sion efficiency and good robustness. Compressive Sensing (CS) base on the signal sparsity can si-multaneously realize sampling and compression, and it saves storage space and computing resourceso as to simplify the encoder. Distributed Compressive Video Sensing (DCVS) is generated bycombining CS theory with DVC. The DCVS fully exploits advantages of CS and DVC, and it isvery appropriate for wireless video transmission with energy and bandwidth constraint and has in-creasingly become the focus of researchers attention in video coding domain.Aim at the existing problems in wireless video communication system, In this thesis, the re-search topic is Distributed Compressive Video Sensing, and the purpose of research is to improvethe rate-distortion performance of video on the premise that the low complexity of encoder can beguaranteed, and several key technologies in DCVS can be studied deeply and acquired some re-search results in the following four aspects:1. For the sparse representation of video in DCVS system, Principle Components Analysis(PCA) is proposed to fully exploit temporal-spatial correlation of video sequences for realizing theadaptive and dynamic sparse representation. Besides, in the basis of the local and remote sparsemodel, the non-local sparse model is proposed to improve the codec performance of DCVS system.2. For the video reconstruction in DCVS system, the existing CS recovery algorithms are com-pared in terms of rate-distortion performance and decoding complexity. Besides, the hybrid prioriknowledge which consist of video sparsity and the non-local similarity is used to establish a DCVSreconstruction algorithm. The proposed algorithm can improve the reconstructed video quality inthe DCVS system at the cost of some certain decoding complexity. 3. For the allocation of measurement rate and dictionary training in DCVS system, the strategyto allocate adaptively measurement rate and the method to train the dictionary are proposed to im-prove the rate-distortion performance of DCVS system. At the encoder, the key frames are meas-ured at the high measurement rate, and the non-key frames are measured by using block CS de-pending on the adaptive allocation of measurement rate. At the decoder, firstly, the decoded keyframes and the adaptive allocation are used to dynamically allocate the measurement of each blockin the non-key frames, and the information on measurement rate allocation is transmitted to the en-coder by the feedback channel; Then, the side information is generated by using the decoded keyframes, and the best dictionary will be trained adaptively; Finally, the CS recovery algorithm is usedto reconstruct the video.4. In order to verify the effectiveness of the DCVS system, the mobile video communicationsystem is designed. In this system, the mobile devices use the CS-based video encoder owning lowcomputational burden and energy consumption to produce the bit stream, and the CS-based videodecoder reconstruct the video signal by using the output bit stream while the reconstructed videowill be encoded again as bit stream by the H.264/MPEG video encoder. The bit stream is transmit-ted to mobile devices for realizing the low decoding complexity.