DCVS中插值多假设预测重构算法

摘要

结合视频编码系统中分像素精度运动估计的思想,对分布式视频压缩感知系统 (DCVS)中插值多假设预测重构算法进行研究.根据自然图像像素连续特性,在传统帧内多假设集合的基础上,对关键帧进行插值构成字典,进行帧内插值多假设预测重构;根据视频序列中物体在帧间产生的位移量具有任意性的特点,在传统帧间多假设集合的基础上,对压缩感知帧进行插值构成字典,进行帧间插值多假设预测重构.实验结果表明,与多假设预测重构算法相比,该算法使视频序列重构质量提高了2 dB-4 dB.%Coupling the concept of fractional-pixel interpolation with MH-CS,a fractional-pixel interpolation multihypothesis prediction (FPI- MH-CS)algorithm for distributed compressive video sampling (DCVS)was studied.For key frames,according to the continuity of natural image pixel,the dictionary for block-based intra-frame FPI-MH-CS was built by interpolating the hy-pothesis set of traditional intra-frame MH-CS.For CS frames,according to the randomicity of the displacement of moving objects between successive frames for video sequence,the dictionary for block-based inter-frame FPI-MH-CS was built by interpolating the hypothesis set of traditional inter-frame MH-CS.Experimental results demonstrate that compared with the MH-CS algo-rithm,the proposed method can improve the PSNR of sequences by 2 dB-4 dB.