摘要
匹配追踪是一种直接求解稀疏问题的有效方法,但是匹配追踪的完全搜索方案耗费大量计算时间.已有的一些避免这种完全搜索的方案对贪婪的学习字典并不适用.本文提出一种基于索引字典的正交匹配追踪(OMPID)的新方法,通过改进的聚类方法建立学习字典的索引,通过索引寻找最佳匹配原子,从而大大地减少了蛮力搜索的时间开销.实验表明,本文方法构造的字典能够极大地提高算法的时间性能,同时对图像的降质影响不大.理论和实验分析还对OMPID算法的相关的参数设置提供了建议.
Matching pursuit is an effective method of solving the sparse problem,but it＇s full search scheme leads to too much CPU time overhead.Some existing methods,which avoided the brute force search,cannot apply to MP based on learned dictionary because these methods exploited some characteristics of atoms.This paper presents a novel method-Orthogonal Matching Pursuit based on Indexed Dictionary（OMPID）.OMPID first indexes the learned dictionary by utilizing improved clustering and then find the most matching atom via the index of learned dictionary.Thus,it avoids time overhead which the full scheme brings about.Our experiments demonstrated that OMPID can expedite the execution of the algorithm immensely under the condition of trying to keep the quality of image denoising when the size of learned dictionary is big.What＇s more,this paper made some experimental suggestions on the related parameters of OMPID.
出处
《小型微型计算机系统》
CSCD
北大核心
2011年第6期1103-1107,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(60772091)资助
关键词
匹配追踪
学习字典
稀疏表示
索引
matching pursuit
learned dictionary
sparse representation
index
