一种分块图像的BP压缩感知重构算法  

A Kind of BP Compressive Sensing Reconstruction Algrithim Based on Image Segment

作　　者：刘继忠[1] 郑恩涛 贺艳涛 付珊珊 赵鹏 LIU Jizhong;ZHENG Entao;HE Yantao;FU Shanshan;ZHAO Peng(College of Mechatronics Engineering,Nanchang University,Nanchang,Jiangxi 330031,China)

机构地区：[1]南昌大学机电工程学院,江西南昌330031

出　　处：《新疆大学学报（自然科学版）》2018年第3期340-347,共8页Journal of Xinjiang University(Natural Science Edition)

基　　金：国家自然科学基金(61273282);江西省教育厅科学基金(GJJ12005)

摘　　要：在图像压缩感知重构中,针对重构效果和耗时不兼得的问题进行深入研究.基于离散余弦基稀疏表示,选用随机高斯矩阵进行观测采样,针对基追踪(BP)重构算法精度相对较高同时计算复杂度也高的特点,结合图像分块可以提高运算速度和精度这一优点,提出一种基于分块图像的基追踪(BP)重构算法,并与常用的正交匹配追踪OMP算法、BP算法、COSAMP算法、基于分块图像的压缩采样匹配追踪(COSAMP)算法、基于过完备字典(KSVD)的OMP重构算法和基于过完备字典(KSVD)的BP重构算法进行对比;借助MATLAB进行仿真实验,得到不同采样率下的重构图像以及重构图像的峰值信噪比和运行时间.实验结果表明:基于分块图像的基追踪(BP)重构算法不但峰值信噪比(PSNR)比普通算法高出1～10d B不等,而且运行时间比较短,所以本文所提算法兼顾了重构精度和运算效率.另外,对本文所提算法分块大小、稀疏度设置多大为最优这两个问题进行大量重复实验,最后确定分块大小为8*8、稀疏度设置为图像矩阵(N*N)原维度N的0.2～0.4倍时为最优.In image compressive sensing, we have to take it all into the in-depth study for solving the problem of reconstruction effect and time-consuming. Based on discrete cosine sparse basis, we choose random gaussian matrix as observed sampling, In view of the better reconstructed image but Basis Pursuit reconstruction algorithm perform slowly. combining with the merits of the image block can improve the precision of image, we put forward a kind of Basis Pursuit reconstruction algorithm based on image block, then the traditional OMP algorithm、Basis Pursuit algorithm、Compressive sampling algorithm、Compressive sampling algorithm based on image block, OMP algorithm and BP algorithm based on Overcomplete Dictionaries for Sparse Representation are compared. Under different sampling rate we perform MATLAB simulation experiment, and we get the reconstructed image with the peak signal to noise ratio of the reconstructed image and the consuming time. The experiment results show that not only the Basis Pursuit algorithm based on image block is 1 to 10 d B higher at the peak signal-to-noise ratio, but also is shorter at the running time than others. Thus the Basis Pursuit algorithm based on image block is the best one, what＇s more,considering the question that how to set the value of image block is a good choice, lots of same experiments are done, getting the solution that every block is setted as 8＊8 and sparsity that is equal to（0.15～0.4） of the old dimensions are the better choices.

关 键 词：图像处理 BP算法 分块 离散余弦变换 COSAMP算法 过完备字典 

分 类 号：TN919.8[电子电信—通信与信息系统]