Technologies On Distributed Video Coding

In conventional video coding framework, e.g. MPEG/H.26X, the encoder is much more complex than the decoder due to heavy computations in motion estimation. This asymmetry in complexity fits well scenarios where a video signal needs to be compressed once but decoded many times, e.g. broadcast or streaming video on demand. However for some scenarios, for example, in wireless low-power video surveillance or wireless video camera systems, the encoders have to be simple and power restricted, but the central node, e.g. the server of surveillance center, has high computing capability and infinite power. At such circumstances a coding approach with simple encoder is desired. Distributed video coding which shifts the complexity from encoder to decoder, is just for these scenarios.The key technologies of distributed video coding are researched in this paper and some novel algorithms are presented. The main contents of this paper are as following:For video surveillance scenarios, this paper presents a new distributed video coding architecture. In our proposed architecture, the reliability of each non-overlapped block in side information is first determined and the reliability matrix of the whole side information is feedback to the encoder. The higher the reliability is the higher the dependence between the block in side information and the same located block in WZ frame is. In WZ frames, the blocks with high reliable side information are not be encoded, and at the decoder the same located blocks in the side information are treated as the reconstructed WZ blocks. Other blocks with low reliable side information are intra-block encoded and are transmitted to the decoder after Huffman encoding. At the decoder, the blocks with low reliable side information are reconstructed using the side information and the received data. Experimental results indicate that, compared with IST-PDWZ, the proposed distributed video coding architecture only needs one-time feedback to decoding one WZ frame, and the decoder complexity is reduced significantly. The mean PSNR value of reconstructed frames is also improved about 1dB compared with IST-PDWZ.To reduce feedback times of distributed video coding, this paper proposes a new encoder rate estimation method. First, the simple three-step motion estimation is introduced into the encoder to estimate the side information of the decoder, and then the number of parity bits the decoder needs for each bit plane is estimated at the encoder according to the Hamming distance of the estimated side information and the current Wyner-Ziv frame. Experiment results indicate that the accuracy of the proposed method is 5.18%-52.93% higher than that of the method proposed by Morbee. And the system performance is also improved to some extent.In the existing distributed compressive video sensing architectures, the researchers reduced the needed number of measurements by exploiting the correlation between successive video frames. However, this is still a difficulty about how to compress the measurements. In this paper, we first propose a Gaussian distribution model to describe the correlations of measurements between a key/CS frame and its side information. Based on this model the measurements of a key/CS frame can be compressed using a channel coder, similar to that in DVC. And then we propose a novel approach on distributed compressive video sensing. Experimental results indicate that the measurement compression ratio of the proposed compression scheme achieves 48%-88%. And the proposed distributed compressive video sensing approach out-performs the scheme without measurement compression about 1-4dB.In our proposed distributed compressive video sensing approach, in order to reduce the system delay caused by the iterative channel decoding of each bit-plane, we further propose a binary symmetrical channel (BSC) model to describe the relationship between the measurement bit-planes of a key/CS frame and its side information at the condition that the scrambled block Hadamard ensemble (SBHE) matrix is used as the measurement matrix. Based on the BSC model, we can obtain the rate-entropy curves of LDPCA codes by experiments in advance. According to the rate-entropy curves, we propose a simple encoder parity-bit rate estimation method. Experimental results indicate that the proposed rate estimation method can reduce 82.35%-95.91% feedback requests with an increase of bit-rate, caused by over estimation, only of 0%-0.98% per frame.