基于余弦相似度的改进ORB匹配算法

针对ORB(oriented FAST and rotated BRIEF)在匹配多相似区域图像时误匹配率高的问题,提出一种基于余弦相似度的改进ORB算法。该算法首先通过汉明距离(Hamming distance)最近邻匹配,计算匹配特征向量的余弦相似度;其次通过梯度计算法计算余弦相似度的最优阈值范围;然后将汉明距离高自由度最近次近邻粗匹配结果,利用余弦相似度不变性剔除不在最优阈值范围的匹配点;最后用RANSAC算法再次精确匹配。实验结果表明:该算法能在保证匹配特征点数以及实时性的基础上,高效降低误匹配率,尤其在匹配多相似区域图像时,可将原算法产生的误匹配率降低80%左右,同时该算法还超越了原算法在图像发生视角、旋转、尺度、模糊、光照变化时的适应性。

An improved ORB matching algorithm based on cosine similarity

An improved oriented FAST and rotated BRIEF(ORB)algorithm based on cosine similarity is proposed to solve the problem of high mismatch rate when ORB matches images with multiple similar regions.Firstly,the cosine similarity of feature vector is calculated by matching the nearest neighbor of Hamming distance.Secondly,the optimal threshold range of cosine similarity is calculated by gradient algorithm.Then,based on the coarse matching of the nearest neighbor of Hamming distance with high degree of freedom,the matching points that are not in the optimal threshold range are eliminated by cosine similarity invariance.Finally,RANSAC algorithm is used in sub exact match.The experimental results show that the algorithm can effectively reduce the error matching rate on the basis of ensuring the number of matching feature points and real-time performance,especially when matching images with multiple similar regions,it can reduce the error matching rate generated by the original algorithm by about 80%.At the same time,compared with the original algorithm,the proposed algorithm has better adaptability when the angle of view,rotation,scale,blur and illumination change.