多尺度自卷积变换与MSER组合的影像匹配方法研究  

Research on image matching method based on the combination of multi-scale self-convolution transform and MSER

作　　者：吉长东[1] 沈晓刚 JI Changdong;SHEN Xiaogang(School of Surveying, Mapping and Geographical Sciences, Liaoning Technical University, Fuxin 123000, China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000

出　　处：《测绘工程》2021年第5期23-31,共9页Engineering of Surveying and Mapping

摘　　要：针对影像仿射变换较大,目前常用局部特征提取算法匹配效果不理想的问题,提出以信息熵为约束条件,筛选出高信息量的MSER特征区域。文中利用多尺度自卷积(MSA)变换的灰度概率密度函数直方图熵构造新的特征向量,并以此作为优化后的MSER局部特征区域的描述符,组合成一种新的具有抗仿射变换的局部特征匹配方法,经过两组具有仿射变换的近景影像对该算法的实验验证,算法整体上可以有效地剔除10%以上的质量不好的特征区域。研究表明,在匹配准确性上,效果要优于MSER算法,尤其是程度较大的仿射变换,平均匹配准确率也要比MSER高。In view of the large affine transformation of images and the unsatisfactory matching effect of the commonly used local feature extraction algorithms,the information entropy is proposed as a constraint condition to filter out the MSER feature regions with high information content.This paper constructs a new feature vector using the histogram entropy of the gray-scale probability density function of the multi-scale auto-convolution transformation,and uses it as the descriptor of the optimized MSER local feature region to form a new local feature with anti-affine transformation matching method.After two sets of close-range images with affine transformation are used to verify the algorithm,it is concluded that more than 10%of poor-quality feature areas can be effectively eliminated as a whole.Studies have shown that the matching accuracy is better than the MSER algorithm,especially the greater degree of affine transformation for which the average matching accuracy rate is also higher than that of MSER.

关 键 词：多尺度自卷积 MSER 仿射不变 特征匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]