Research On Prediction Algorithms For Compressed Video Sensing

For traditional video coding,video signal is sampled based on the Nyquist sampling theory with sampling frequency greater than or equal to twice the maximum frequency of the signal,then the spatial and temporal redundancy information in the video signal is removed by the conventional encoding method.As people's requirements on the quality of multimedia content getting higher and higher,the burden at the video encoder is getting heavier.But the traditional video coding method is not suitable for the application environments(eg wireless video surveillance)with limit in power consumption,storage capacity,and computing power.Compressed sensing(CS)conducts sampling and compression simultaneously,saving enormous sampling resources while reducing the sampling complexity significantly.Thus is suitable for the application scenarios with a resource-deprived sampling side.Compressive sensing based distributed video coding attracts considerable attention,in which,making full use of correlation between frames to reconstruct video efficiently has become one of the main research area.This thesis focuses on the prediction algorithms in compressive video sensing(CVS).The major work is including two parts as follows:1.Current Multi-Hypothesis prediction(MH)algorithms used in CVS have the disadvantages of high computational complexity and low prediction accuracy when dealing with fast moving sequences.Besides,MH in measurement domain just employ the sum of absolute difference(SAD)principle to select hypothesis blocks,which usually introduce noise in the prediction blocks and decrease the reconstruction quality for neglecting the one-to-many relationship between the given measurement and original signals.To address these issues,this paper takes advantage of the motion features in video and proposes a Multi-Hypothesis prediction scheme based on fast diamond search with two matching regions(MH-DS).Moreover,a new matching criterion integrating Mean Square Error(MMSE)with Maximum Pixels Counting(MPC)is proposed in MH-DS in order to get more relevant hypothesis blocks.Simulation results show that the proposed MH-DS reduces the prediction complexity while maintaining large search area,and obtain higher prediction accuracy than the state-of-the-art CVS prediction methods.2.The existing multi-hypothesis prediction algorithm(including the first work in this thesis)is carried out under the condition that video is partitioned into blocks with same size.In which two shortcomings exists:(1)For the block with complicated movement,low prediction accuracy is unavoidable;(2)For the smooth motion region,the motion vectors of adjacent blocks are very similar,searching the best matching block for each block causes large algorithm complexity.To address these issues,this paper proposes Hierarchical MultipleHypothesis(Hi-MH)idea,which uses different block matching methods for different moving area.And an implementation method is proposed,on one hand,the motion vector of the current block is predicted by that of its neighboring blocks in the smooth motion region to reduce computational complexity,on the other hand,block matching is implemented based on smaller blocks,or prediction is carried out based on auto-regressive model in the complicated moving area to obtain higher predict accuracy.Simulation results show that the proposed Hi-MH reduces the computational complexity for video sequences or motion regions with simple movement,while improving the predict accuracy for complicated movement sequences.