关键词: 压缩感知, 视频重构, 多参考帧, 多假设预测, 量化

Abstract:

The existing multi-hypothesis prediction methods for compressed video sensing ( CVS) select all possible blocks within the search space of reference frames as the hypotheses,which causes a high computation load in solving linear weighting coefficients and impairs prediction accuracy.To address this issue,a multi-reference framesbased optimal multi-hypothesis prediction algorithm for CVS reconstruction is proposed in this paper.In the algorithm,first,those search blocks which have the smallest sum of absolute differences ( SAD) from current block in measurement domain are selected from multi-reference frames as the optimal hypotheses of current block.Then,the hypotheses are weighted both linearly and adaptively to fully excavate the temporal correlation between video frames.Thus,the prediction accuracy is improved and the computation load in solving linear weighting coefficients is reduced.Finally,the compressed sensing measurements are quantized through the frame-based DPCM quantization to improve video compression efficiency and rate-distortion performance.Simulation results show that,in comparison with the existing CVS reconstruction algorithms,the proposed algorithm achieves higher video reconstruction quality.

Key words: compressed sensing, video reconstruction, multi-reference frames, multi-hypothesis prediction, quantization