Research Of Distributed Video Coding Based On Compressed Sensing

With the wide application of mobile video technology, compression and transmission of video on portable devices with the limit of power,computing and storage become the trends of multimedia information technology.Distributed video coding(DVC) with simple encoding and complexity decoding come into being. DVC which has"independent encoding,joint decoding"characteristics based on distributed source coding(DSC),and it make full use of the relevant characteristics of the video to simplify the process of coding system. In recent years,in order to reduce of the complexity of the coding system and expand the application of the DVC technology, compressed sensing(CS) which can compressed signal directly in the transform domain has received extensive attention in the filed of DVC.In this paper,in order to fully exploit the characteristic of inner video and inter video,by integrating DVC and CS, a new distributed compressed video sensing(DCVS) based on multi-hypothesis prediction is proposed.In the scheme,the video is divided into key-frame and non-key frame,at the encoder two kinds of frame are both measured and key-frames has a higher measurement rate;at the decoder,key-frame has CS reconstruction,while non-key frames is reconstructed through using the correlation of the video,which is also the key point in this paper.In the reconstruction of non-key frames, prediction frames play an important role, we extend the concept of the prediction plus residual reconstruction to the multi-hypothesis prediction,and the prediction frames is generated by using modified multi-hypothesis prediction.In our multi-hypothesis prediction,the adjacent frame which has reconstructed and the estimated of the current frame are regarded as the hypothesis,and linearly combines them to generate a new prediction,the size of matching block and search window of the motion estimation are adjusted in the iterative process of residual reconstruction,so the video reconstruction has better performance through updating the multi-hypothesis prediction.In this paper, we study the theory of DVC firstly and focus on the generation algorithm of the side information.Next we study the theory of CS,for the picture we use compressed sensing based on block,and compare the performance of several reconstruction algorithms by simulating.Then we study the distributed compressed sensing(DCS) and focus on the JSM-1 model which is widely used in the distributed video,the simulation results show that the JSM-1 model can achieve better performance using fewer measurement than independent CS reconstruction.Then,the modified multi-hypothesis prediction is applied to the distributed compressed video sensing.Finally,the simulations have been carried out to demonstrate that the proposed scheme can improve the performance of reconstruction efficiently.