基于动态多模式匹配的视频压缩感知三步重构

Three-phase video compressive sensing reconstruction viadynamic multi-pattern matching

摘要:

相较于奈奎斯特-香农定理所要求的采样数据量,压缩感知理论表明采用较少的测量值就可以实现高维信号的重构,因此压缩感知在视频信号传感中具有很大的潜力.现有的视频压缩感知重构算法是利用多假设预测来得到残差模型,大量文献采用基于最小均方误差的方法挑选多假设匹配块,由此对视频信号进行重构,然而没有考虑最大化重构视频的整体结构相似性,在图像重构的整体质量效果上存在较大改进空间,并且挑选匹配块的模式没有采用自适应的选择机制,挑选匹配块的方式较为单一.通过增加一定的复杂度,本文提出了一种基于动态多模式匹配的视频压缩感知三步重构算法,该算法主要包括三大步骤:第一步,对压缩感知的每一视频帧进行独立的重构；第二步,从参考帧中动态地挑选匹配块进行非关键帧的重构；第三步,基于整体结构相似性对重构帧进行最终挑选,完成多帧重构.实验结果表明,所提算法能够在重建端有效提高多假设过程的预测精度,与当前最优的视频重构算法相比,进一步提升了重构质量.

Abstract:

Compressive sensing theory indicates that high-dimensional signal reconstruction can be obtained from far fewer measurements than those required by the Nyquist-Shannon sampling theorem, and compressive sensing has a great potential in video signal sensing. The existing reconstruction algorithms utilize the multihypothesis prediction to derive the residual model. A large number of studies adopt the method based on the least mean square error to select multiple hypothesis matching patehes and reconstruct the video, while the maximization of the structural similarity is not considered in these algorithms, and there is much room for improvement in the overall quality effect of image reconstruction. Therefore, a three-phase video compressive sensing reconstruction algorithm is proposed in this paper on the basis of dynamic multi-pattern matching, in which the first phase reconstructs each frame independently, the second phase dynamically selects the hypothesis patches from the reference frames and reconstructs the frames, and the final reconstruction result is obtained in the third phase with the best structural similarity. Experimental results demonstrate that compared with the state-of-the-art algorithm, the proposed algorithm could obtain better prediction accuracy and reconstruction quality for video compressive sensing.