基于网格运动统计的自适应图像特征匹配算法

Self-adaptive image feature matching algorithm based on grid motion statistics

针对网格运动统计(GMS)算法性能依赖特征点数量且当特征点检测较少时存在误匹配集中的问题,结合一致性约束思想,提出了一种基于网格运动统计的自适应特征匹配算法.首先对待检测图像引入网格划分,依次对每个网格区域设置自适应阈值并进行特征点检测;然后使用旋转特性的二进制描述(rBRIEF)算法对特征点描述并基于汉明距离完成特征点匹配;最后采用GMS算法做初次误匹配点剔除,利用随机抽样一致算法筛选出精确匹配点.实验结果表明:该算法能有效剔除误匹配点,提升匹配质量且实时性高,对于低纹理结构的图像匹配也具有很好的鲁棒性.

In order to solve the problem that the performance of grid motion statistics(GMS) algorithm depends on the number of feature points and there exists mismatching concentration when feature points are detected less,an adaptive feature matching algorithm based on grid motion statistics was proposed with the idea of consistency constraints.By introducing grid division into the detection image,the self-adaptive threshold was set for each grid region to detect feature points in turn.Then,the rotated BRIEF(rBRIEF) was used to describe feature points and feature point matching was completed based on hamming distance.Finally,the GMS algorithm was adopted to eliminate the initial mismatching,and random sample consensus algorithm was used to select the accurate matching points.Experimental results show that the algorithm can effectively eliminate mismatching points,and improve matching quality with high real-time performance,and also has good robustness for image matching with low texture structure.