Research On Key Technologies Of Distributed Video Compression

With the development of mobile acknowledge e internet and information technology, more and more mobile video applications are widely used in many aspects of society. However, in the conventional video coding standards, such as MPEG/H.26X, the video encoder needs to make use of the video statistical character and explorer complex motion estimation and compensation solutions for much better rate-distortion (RD) performance. As a result, the computing power of the encoder is typically5to10times more complex than that of the decoder. Therefore, it is unrealistic to the mobile terminals with limit computational capability and energy, and increases the cost for the mobile terminals. In addition, the utilization of prediction frames in the conventional video standards also results in weak resisting-error capability in wireless networks and can not fulfill the requirements for video with high quality. Therefore, it is emergent to design a new video coding architecture with a low-complex encoder and robust transmission capacity to replace the conventional video standards.Considering these requirements, distributed video compression technologies with low complexity and robustness become a research topic. It could get rid of the prediction frame at the encoder and shift the computing complexity from the encoder to the decoder by separately encoding and jointly decoding the correlated video frames. Therefore, the research work on distributed video compression in this paper will include two aspects:Distributed Video Coding (DVC) and Distributed Compressive Video sensing (DCVS). The specific research contents are shown as follows:1. In the DVC, realize the side information (SI) revision and the correlation noise model (CNM) parameter revision by using the partially decoded information. Because the decoded information in the video frame could partially represent the video frame character, on one hand we propose to partition the frame into several parts. The encoder will encode every part sequentially and utilize the decoded parts as prior information to revise the SI and CNM parameter; on the other hand we propose to make use of the decoded bitplanes as prior acknowledge to revise the SI the CNM parameter. According to the simulation results, the proposed SI revision method and the CNM parameter method could improve the RD performance.2. In the DVC, realize an improved decoding scheme in our proposed DVC architecture. Considering the movement of the objects in the SI frame, in this scheme all of blocks in the frame are classified as three modes:Zero Mode (ZM), Classical Mode (CM) and Revise bit order Mode (RM). At the encoder, according to the block mode information, the quantified coefficients are converted to their binary representations with different methods. At the decoder, according to the block mode information, different bits belonging to different block mode in one bitplane will have different likelihood-ratio calculation methods. And the reconstruction unit also utilizes the block mode information to realize the frame reconstruction. Simulation results show that our proposed decoding scheme could achieve better performance than the traditional decoding algorithm.3. In the DVC, propose a rate estimation method at the decoder. In order to reduce the feedback times and the computing complexity at the decoder, we propose a new decoder rate estimation method. By combining the error estimation code used in the modulation mode selection with the Low-Density Parity-Check (LDPC) matrix in the DVC, this method makes use of the parity bits to estimate the bit error ratio(BER) of the bitplane without increasing any redundant bits. Finally, with the BER the minimum rate of the bitplane is calculated with the conditional entropy. Simulation results indicate that our proposed rate estimation method could save70%decoding time at the maximum.4. In the DVC, according to the importance of data in the video frame, design an LDPC matrix with unequal error protection (UEP) function. In the irregular LDPC code, according to the wave effect in the belief propagation decoding process, the variable nodes having larger degree could get more reference information from the check node connected to it and could be decoded quickly and accurately. Thus, in this paper, we utilize the Progressive Edge-Growth (PEG) algorithm to generate an Weight-Increasing Parity-Check (WIPC) matrix as the LDPC matrix。Simultaneously, the decoder exploit the same method to partition all of blocks in the frame into three mode:Fast Motion (FM)、Moderate Motion (MM) and Stationary/Slow Motion (SM). The encoder will map the bits belonging to different modes to variable nodes with different degrees in the LDPC matrix for UEP. If the pixels located in the block belonging to FM, the corresponding bits of the pixels will be mapped to the variable nodes with the larger degrees/weights in the WIPC matrix; if the pixels located in the block belonging to SM, the corresponding binary representations of the pixels will be mapped to the variable nodes with the smaller weights; otherwise, the remaining binary representations of the pixels will be mapped to the other variable nodes. With this method, we can realize the UEP and obtain better compression efficiency and decoding performance.5. In the DCVS, according to the sparsity of the video signal in the transform domain, design a Low-Density Sensing matrix (LDSM) with UEP function. In this paper, we extend the design thought about LDPC matrix with UEP function in DVC to the DCVS and design an LDSM with UEP function. In this paper, we also adopt the PEG algorithm to generate an WIPC matrix as the LDSM. Then the coefficients after discrete cosine transform will be mapped to the variable nodes with different degree in the LDSM. The larger coefficients will correspond to the variable nodes with larger degree and the smaller coefficients will correspond to the variable nodes with smaller degree. With this method, the larger coefficients could be effectively protected and have better recovery quality at low signal to noise ratios.6. Design a new modulation method with UEP function for DVC Taken a QAM constellation map with Gray code as an example, according to the calculating results, different bit position of the symbol in the constellation map have different priority. Therefore, if we could place the parity bits of more important bitplanes in more protection bit position of the constellation map and place the parity bits of less important bitplanes in less protection bit position, we will realize the UEP of the data bits without consuming any additional resources. The simulation results show that compared with the traditional method in the DVC, PSNR gain for our proposed scheme can reach up to1db.