基于像素相关的图像/视频压缩感知观测矩阵

Correlation of Pixels-Based Measurement Matrix for Compressed Image/Video Sensing

观测矩阵的构造是图像/视频压缩感知中的重要问题之一.为了解决随机性观测矩阵的不确定性以及在信道传输上的压力和提升编码端重构质量,文中提出了一种应用于图像和视频压缩感知的基于邻近像素相关性的确定性稀疏观测矩阵(APM).首先利用文中提出的环形目标点选择法,在观测图像块中均匀地选取目标点;然后利用高斯分布概率密度函数给目标点周围的邻近像素对应的观测矩阵中的元素分配权值,使得每个观测值中仅包含一个目标点和其相邻像素的信息.利用文中提出的APM观测矩阵得到的观测值对图像或视频进行重构时,既可利用相邻像素的相关性,又避免了远距离像素信息的干扰,可得到较高的重构质量.仿真实验表明,与经典和目前文献中较新的几种压缩感知观测矩阵相比,文中提出的观测矩阵具有更高的图像/视频重构质量,较低的重构复杂度,且适用范围更广.

The construction of measurement matrixes is one of the most significant issues in the compressed image/video sensing(CIS/CVS).In order to overcome the uncertainty of random measurement matrixes and the heavy pressure in transmission and to improve the reconstruction quality in encoder ends,a novel deterministic sparse measurement matrix named adjacent-pixel-based matrix(APM)is constructed for the CIS/CVS,based on the correlation of adjacent pixels.During the APM construction,first,an annulus choosing method is proposed to uniformly select the target points in the measured image block.Then,the Gaussian probability density function is adopted to distribute weights to the elements of the measurement matrix corresponding to the adjacent pixels of the target points,so as to ensure that each measurement only contains the information of one target point and its adjacent pixels.Applying the measurement from the proposed APM in the image or video reconstruction helps take advantage of the correlation between adjacent pixels and avoid the interference from the other further pixels,and thus a higher reconstruction quality can be achieved.Simulation results show that,in comparison with other classical and novel measurement matrixes in the literature,the proposed APM achieves higher image/video reconstruction quality with lower reconstruction complexity,and that it is of a wider range of application.