Research On Frame Content And Position Based Distributed Compressed Video Sensing Scheme

Compressed sensing technology breaks through the limitation that the sampling rate must be more than twice of the maximum frequency of the signal in Nyquist sampling theorem.The signal sampling and compression could be realized synchronously in compression sensing,which improves the compression encoding efficiency significantly.Research on compressed sensing theory based video codec schemes has obtained many results.The multihypothesis prediction based video reconstruction becomes the focus of the research due to its perfect performance.However,current video codec schemes usually select the key frame as the reference frame.The correlation among the non-key frames is usually ignored.Besides,the effect of the non-key frame's position on the decoding quality is not considered in current video codec system,the quality of reconstructed non-key frame is unsatisfactory.To solve the above defects in current codec schemes,a novel distributed compressed video sensing scheme based on the content and the position of the video frame is proposed in this thesis.The varying inter-frame correlation among the video frames and the effect of non-key frames' position on reconstruction are applied in our proposal to optimize the data processing in encoder and decoder.The high-quality video compressed sensing reconstruction under the low sampling rate is realized in this thesis.Firstly,this thesis proposes an optimized compressed sensing encoding scheme by taking the high temporal correlation of the video frames into consideration.In our proposal,the image blocks in non-key frames with high similarity to the reference frames are marked as non-encoded blocks.These redundant blocks are not encoded in encoder,which effectively improves the compression encoding efficiency and relieves the transmission burden.Secondly,for the problem of unbalanced non-key frames reconstruction quality,this thesis proposes a video frame position-sensing multihypothesis prediction reconstruction algorithm.The proposal could provide more multihypothesis information source i.e.,the reference frames,for some non-key frames that are far from key frames.In our proposal,the video frames to be reconstructed could get more high-quality reference information,which improves the accuracy of the side information prediction and the quality of the video reconstruction significantly.The proposal effectively avoids the fluctuation of the non-key frames' reconstruction quality and improves the experience of the users especially for the high motion video sequences.Compared with the existing distributed compressed video sensing schemes,the simulation results show that our proposal effectively reduces the amount of the data to be measured.Moreover,under the low sampling rate,the efficiency and the quality of the reconstruction in our proposal are improved simultaneously.These results indicate that the proposed scheme is effective and feasible.