Research On Key Technologies Of Image And Video Compression Based On Distributed Coding

In recent years, some new appearing applications, for example, sensor network, video surveillance system, multispectral image, etc., have being bringing new challenges to image and video compression techniques, i.e., the limited resources of terminals require to minimize the energy consumption and communication traffic at the encoder. The current image/video coding standards, like JPEG, JPEG 2000, MPEG, and H.26x, do not meet the requirements of new applications. With such challenges comes distribted source coding (DSC) technique, whose theoretical basis are Slepian-Wolf theory and Wyner-Ziv theory, which prove that correlated sources can be compressed by separate encoders without performance loss as long as they are jointly decoded. As a fire-new compression technique, DSC can provide a lot of promising performance and supply a gap to the present techniques, so it has attracted a lot of attentions and become a very hot research area. Therefore, research on image and video compression techniques based on DSC are of both theoretical significance and practical valueBased on the present results of DSC, this thesis investigates a number of key techniques of DSC for camera sensor network, video and multispectral image. The main contents and novelties of this thesis are as follows:1. This thesis proposes a pixel domain Markov model based distributed image coding shceme.The proposed scheme compresses the image in pixel domain without exploiting data correlation at the encoder. Markov model is built at the decoder to estimate the spatial correlation of the image and integrated with LDPC decoding so as to remove both intra-image redundancy and inter-image redundancy simultaneously. This scheme not only acquires high compression efficiency but also minizes the amount of compution and communication at each sensor node so that it is albe to utilize the resources in an optimization way and prolong the life of the network.2. This thesis proposes a transform domain classification based distributed video coding scheme.The proposed scheme classifies the blocks of Wyner-Ziv frames and compresses blocks of different types discriminatingly. The classification information is used to assign rate roughly at the encoder and aid the motion estimation at the decoder. The decoder performs bidirectional motion estimation, mode selection and bidirectional motion compensation to generate high quality side information. Compared with the present scheme, this scheme is more efficient in exploitation of spatially varying temporal correlation and thus, improves the compression efficiency.3. This thesis proposes a pixel domain spatial-temporal Markov random field (STMRF) model based distributed video coding scheme.The proposed scheme builds a STMRF model at the decoder and through estimating motion field and intensity field, the motion estimation and spatial correlation estimation are combined, which works alternately with LDPC decoding. During the decoding procedure, the temporal correlation and spatial correlation are learned and exploited progressively so as to refine forward motion estimation and thus improve compression efficiency. This scheme is suitable for low-delay systems.4. This thesis proposes a MRF model based distributed multispectral image lossless coding scheme.The proposed scheme compresses each band independently in pixel domain to provide a low-complexity encoder that is very easy to hardware implementation. Therefore, it is suitable for onboard applications. The decoder performs fuzzy clustering and fuzzy prediction to generate side information and builds a MRF model with adaptive parameters to estimates the local statistic of the image, which is integrated with LDPC decoding. This scheme takes full adavantage of the nonstationarity of multispectral images, thus it is albe to remove the spectral redundancy and spatial redundancy in an efficient way.5. This thesis proposes a progressive distributed multispectral image coding scheme.The proposed scheme divides the image into a number of slices, which are encoded independently but decoded in sequence. The decoder performs a region-based adaptive prediction algorithm to generate side information using the decoded data. This scheme enables progressive transmission and improves the compression efficiency of the slices progressively. With the characteristics of low-complexity encoding, high compression efficiency, progressive transmission and lossless reconstruction, this scheme is very suitable for onboard applications.This thesis designs a lot of experiments to evaluate performance of the proposed schemes and the experimental results show that, the proposed distributed image/video coding schemes can improve compression efficiency while providing a very low-complexity encoder. The proposed key techniques make some theoretical breakthroughs and technical innovations.