基于图像外形的多尺度相关性检索算法  

Multiscale Relativity Algorithm for Shape-Based Image Retrieval

作　　者：黄宝娟[1] 柯芬蓉[2] 庄健[1] 徐仁鹏[1] 于德弘[1] 

机构地区：[1]西安交通大学现代设计及转子轴承系统教育部重点实验室,西安710049 [2]西安理工大学理学院,西安710048

出　　处：《西安交通大学学报》2007年第1期69-72,81,共5页Journal of Xi'an Jiaotong University

摘　　要：采用数学分析的方法比较了图像检索算法中常用的2种坐标系,发现极坐标更加有利于图像的多尺度处理.设计的边缘序列点的插值算法,既保证了2个比较序列的长度相同,又保留了序列中所有的边缘特征点.根据极坐标下边缘序列点的插值算法和相似轮廓在空间距离上相关性最大的特点,给出了多尺度相关性的检索算法(MSRA),该算法具有对图像尺度变化不敏感,而对图像轮廓变化敏感的特性.通过对自然类图像库和人工类图像库200幅图像的检索,表明该算法的性能高于常用的图像检索算法,与笛卡儿坐标下普通边缘点傅里叶搜索算法(CBPFD)相比,MSRA的搜索识别率在自然类图像库中高于CBPFD的29%,而在人工类图像库中几乎是CBPFD的2.5倍.Two kinds of image retrieval reference frames are compared mathematically, and it is found that the polar coordinates facilitate processing multi-scale images. The interpolation method for edge sequence points is designed, where two compared clusters edge points are endowed with the same length, and all original characteristic boundary points are reserved. By means of the characters of similar image boundaries with the maximum relativity in the space distance, the multi-scale relativity algorithm for shape-based image retrieval （MSRA） is proposed, whieh is sensitive to the changed image contours, but insensitive to the image size. Compared with common boundary points Fourier descriptor algorithm （CBPFD）, the identification rate of MSRA gets higher 29% than that of CBPFD in the natural image collections and the identification rate of MRASB is almost 2.5 times of that of CBPFD in the artificial image collections.

关 键 词：检索算法 相关性 多尺度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]