用稀疏相似性度量求解压缩传感矩阵  

Approach to solving compressive sensing matrix based on sparse similarity measurement

作　　者：石亮[1] 祁云嵩[1] 宋晓宁[1] 

机构地区：[1]江苏科技大学计算机科学与工程学院,江苏镇江212003

出　　处：《计算机应用研究》2014年第10期3192-3195,共4页Application Research of Computers

基　　金：江苏省自然科学基金资助项目(BK2012700);人工智能四川省重点实验室开放基金资助项目(2012RZY02)

摘　　要：提出一种用稀疏相似性度量求解压缩传感矩阵的方法,并将其应用在图像重建和识别领域中。首先构造一种稀疏相似性度量,然后将其嵌入到传感矩阵的模糊代价函数中,最终传感矩阵的原子更新按照模糊方式进行计算。用该方法优化后的观测矩阵与字典矩阵之间保持了低相干性,并且样本的稀疏信号在相同重构条件下具备了更优的测量数目和质量。在ORL和FERET人脸数据库及91幅自然图像库上的实验结果验证了该算法的有效性。This paper proposed an approach to solve compressive sensing matrix based on the sparse similarity measure. First, it designed a measurement of sparse similarity between arbitrary samples. Second, it proposed a new fuzzy cost function for op- timization of sensing matrix by which the update of atoms from sensing matrix were fuzzifized handled sequentially. By this means, it could obtain the low coherence between the properties of observation matrix and dictionary matrix, meanwhile, the merit of the method was that the sparse signal had desirable properties for the number of measurements and representation qua- lities under the same reconstruction conditions. Extensive experimental studies conducted on ORL, FERET face images and 91 natural images databases show that the effectiveness of the proposed method.

关 键 词：稀疏相似度量 压缩传感矩阵 模糊集 图像识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]