图像匹配中的数据存储方法  

Storage method for data in image matching

作　　者：巨新刚[1] 杨靓[1] 秦昳[1] 黄士坦[1] 

机构地区：[1]西安微电子技术研究所,陕西西安710054

出　　处：《武汉大学学报（工学版）》2011年第3期399-403,共5页Engineering Journal of Wuhan University

基　　金：民用航天科研预先研究项目(编号:C5220060318)

摘　　要：在图像匹配中,针对已有图像数据存储方法不能适用于稀疏矩阵的存储,提出了一种可适应紧矩阵和稀疏矩阵的图像数据存储方法,紧矩阵采用逐点存储方法,稀疏矩阵采用行格式存储法(CSR:Compressed Sparse Row).为了满足多种图像匹配算法对数据的存储要求,将存储阵列块划分为8个存储子块和4种存储模式,可存储参考图和模板图.通过存储控制块的控制,图像数据可并行或单独访问,增加了数据读取效率和灵活性.通过分析和比较,该存储方法能有效节省存储空间,改善存储性能.According to the existing storage method for image data inapplicable to sparse matrix in image matching,a new storage method for both tight matrix and sparse matrix is proposed;the point-to-point storage method is for tight matrix,and the compressed sparse row（CSR） format storage method is for sparse matrix.In order to meet a variety of image matching algorithm of data storage requirements,the storage arrays are divided into eight sub-storage blocks and four storage modes;and it can storage both reference image and matching image.By controlling the storage control block,image data can be parallel or separately access so as to increase data reading efficiency and flexibility.By analysis and comparison,it can save storage space efficiently and improve storage performance.

关 键 词：图像匹配 稀疏矩阵 CSR格式 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]