运动平滑与核线约束的特征匹配提纯算法

针对基于最近邻距离比率约束和几何条件约束的影像特征匹配方法,在处理重复纹理影像以及视角和尺度变化大的宽基线影像时,存在匹配正确率低和匹配点对数量少的问题,设计了结合运动平滑约束与核线约束的AKAZE特征匹配提纯算法。①为提高影像特征的不变性、显著性和时效性,采用AKAZE算子提取影像特征,经暴力匹配快速构建初始匹配集;②基于网格运动统计区分真假匹配,剔除不能满足运动平滑约束的匹配点对;③为消除局部相似特征引起的误匹配,采用核线约束提高匹配纯度。实验结果表明:所提算法实现了重复纹理影像、倾斜影像和宽基线影像同名点对的高精度匹配,增加了匹配点对数量,拓展了特征匹配的适用范围。

Feature matching purification algorithm with constraint of motion smoothness and epipolar geometry

While dealing with highly repetitive texture images and wide baseline images with large viewpoint and scale changes,traditional image feature matching methods,based on the constraints of nearest neighbor distance ratio and geometric condition,exist the deficiency of low matching accuracy and few matching points.In view of this situation,an AKAZE feature matching purification algorithm is proposed,which combines motion smoothness constraint and epipolar geometry constraint.Firstly,in order to improve the invariability,distinctness and timeliness of feature extracting,AKAZE operator is used to obtain high-quality and efficient image feature.The initial matching set is quickly constructed by brute-force match.Secondly,Grid-based Motion Statistics is used to eliminate matching pairs,which do not satisfy motion smoothness constraint.Finally,in order to remove the mismatches caused by local similar features,epipolar geometry constraint is utilized to match the features precisely and improve the purity of the matching set.Experimental results show that the proposed algorithm achieves highprecision matching of homonymous point-pairs for repeated texture images,oblique images and wide baseline images,increases the number of feature matching points and expands the application scope of feature matching.