基于稀疏表示的差分图像估计方法研究  

ON DIFFERENCE IMAGE ESTIMATION BASED ON SPARSE REPRESENTATION

作　　者：陈平生[1] 龙丹[2] 

机构地区：[1]绍兴职业技术学院浙江绍兴312000 [2]浙江大学浙江杭州310058

出　　处：《计算机应用与软件》2013年第6期4-7,18,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(30900358/C100701);浙江省教育厅科研项目(Y2011122724);浙江省新世纪高等教育教学改革研究项目(yb09138)

摘　　要：差分图像能够显示目标场景随时间的变化。采用一种基于图像特征的方法,由目标场景的一组压缩测量数据得到差分图像。该方法主要包含两方面的内容:首先设计最优的采样矩阵以得到压缩测量数据;采用闭环的迭代方法得到差分图像的估计值,包括基于l_2和l_1的方法。采用基于l_2范数的方法能够由压缩测量值直接估计出差分图像而不需要首先重建目标场景。这种方法主要利用了目标场景在连续时间点空间和时间上的相关性。基于l_1范数的方法主要采用一种改进的全变差方法和基追踪降噪方法。仿真表明了该方法的有效性。Difference images can show the changes in object scene together with the time. In this paper, we use an image feature-based method to get the difference images from a set of compressive measurement data of the object scene. The method mainly includes two steps. First it designs the optimal sampling matrix for getting compressive measurement data. Secondly, it uses closed-loop iterative technique for estimating the difference images, including the 12 and l1-based techniques. Employing the technique of l2-based paradigm can directly estimate the difference image from compressive measurements without the initial reconstruction of object scene. This method mainly exploits the spatial and temporal correlations of the object scene at two consecutive time instants. The l1-based paradigm technique mainly adopts a kind of modified total variation method and the method of basis pursuit denoising. Simulation demonstrates the efficacy of these techniques.

关 键 词：压缩感知 差分图像 全变差 基追踪 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]