基于三角形区域仿射不变性的特征匹配算法  

A Feature Matching Algorithm Based on the Affine Invariant Triangle Area

作　　者：沈洋[1,2] 熊明磊 单士娟 陈林[1] 

机构地区：[1]宿迁学院信息工程学院,宿迁223800 [2]北京大学工学院,北京100871

出　　处：《科学技术与工程》2017年第25期261-267,共7页Science Technology and Engineering

基　　金：国家自然科学基金(51575005);宿迁市科技创新专项基金(S201510;Z201415;Z201537)资助

摘　　要：针对图像进行匹配时,若图像中存在相似的区域,且图像之间发生了大视角的仿射变化时,会产生大量误匹配的问题。提出了一种基于三角形区域仿射不变性的特征匹配算法:AIT-SIFT;此算法首先构造扇形描述符,利用平均KL距离得到候选匹配特征点集;再通过互惠最佳匹配策略来构造仿射不变量,利用此不变量,对候选匹配点集进行筛选。实验结果显示,AIT-SIFT算法不仅可以提供足够多的正确匹配的特征点对;同时在查全率、精确度方面有明显提高。In the process of matching,if there are similar regions in images,and when one image has a change of large viewing angle,it will produce a large number of false matchings. In order to solve this problem,an AITSIFT algorithm based was proposed on affine invariant triangle area. Firstly,some feature descriptors based on sector area partitioning are created,secondly,these descriptors are filtered by Kullback-Leibler divergence.Thirdly,an affine invariant is generated according to the reciprocal best matching algorithm,the affine invariant is applied to eliminate the errors in matching. Experiments show that the new approach not only provides many corrects in matching,but also has higher recall and precision.

关 键 词：图像匹配 SIFT算法 仿射不变量 KL距离 关键点提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]