基于压缩感知的视频信号采集观测值渐进量化算法

Progressive Quantization Algorithm for Video Signal Acquisition Based on Compressed Sensing

在基于压缩感知的视频信号采集中,观测值的量化方法会对重构质量产生重大的影响.为了设计一种性能较优的观测值量化方法,根据视频信号的帧间相关性和压缩感知的视频采集信号观测值特性,提出了基于压缩感知的视频采集信号观测值渐进量化算法.该算法将非关键帧观测值均匀量化后只传输若干不太重要的码平面,在重构端利用邻近的已解码帧通过运动估计生成该非关键帧的边信息帧,再通过观测得到该非关键帧观测值的估计,结合接收到的不太重要码平面信息,通过渐进量化的逆量化得到精确的观测值.实验结果表明：与均匀量化算法相比,文中算法在不增加编码端复杂度和不降低视频序列重构质量的基础上,能大幅降低码率;在相同码率下,不同序列获得的平均增益在0.5～2.0 d B之间,具有较高的率失真性能.

In the video signal acquisition process on the basis of compressed sensing,the quantization method of measured values influences the reconstructed quality remarkably. In order to design a high-performance quantization method of measured values,a progressive quantization algorithm for compressed video sensing measurements named VPQ（ Video Progress Quantization） is proposed on the basis of the inter-frame correlation of videos and the characteristics of measured compressed video sensing signals. In this algorithm,the measured non-key frames are quantized and only some bitplanes with less importance are transmitted. At the decoder,neighbor reconstructed frames are applied to motion estimation to generate the side information of non-key frames,and then the non-key frame measurements are estimated by measuring the side information frame. Finally,in combination with the bitplanes with less importance transmitted from the encoder, accurate measurements are obtained via the inverse quantization of progressive quantization. Experimental results show that,in comparison with uniform scalar quantization,the proposed VPQ algorithm helps greatly decrease the code rate without additional complexity and reconstruction quality degradation; and that it is of higher rate-distortion performance,with a gain ranging from 0. 5 to2. 0 d B for different sequences.