一种提高SIFT特征匹配效率的方法

摘要

In order to solve this problem,the authors reformed the form of similarity measurement of SIFT feature descriptors by using city-block distance instead of Euclidean distance to decrease the time complexity of the similarity measurement formula.Besides,a hypothesis algorithm about the nearest neighbor and the second-nearest neighbor was proposed,which supposed arbitrary two features in the image to be matched were the nearest neighbor point and the second-nearest neighbor point respectively and these two points can be replaced by comparing the distance of the current feature from other features in the image to be matched and the distance of the current feature from the supposed two features,finally the actual nearest neighbor point and the second-nearest neighbor point were gotten.The algorithm reduces the number of compares of features involved in the process of similarity computation and thereby decreases the amount of the computation of the algorithm.Experiments show that the proposed algorithm improves matching efficiency of SIFT features while keeping robustness unchanged,and which can provide safeguard for those applications with high real-time requirements.%为了提高SIFT特征匹配的效率,首先改造了SIFT特征描述符相似性度量的形式,以街区距离代替欧氏距离作为特征描述符之间的相似性度量,降低了相似性度量公式的时间复杂度;其次,提出了最近邻和次近邻假设算法,即假设待匹配图像中任意2个特征点为最近邻点和次近邻点,通过比较当前特征点与待匹配图像中其他特征点之间的距离,以及当前特征点与假设的最近邻和次近邻之间的距离,实现最近邻和次近邻的替换,最终得到实际的最近邻点和次近邻点。算法减少了相似性计算过程中特征点比较的次数,从而减小了算法的计算量。实验结果表明,提出的算法在保持鲁棒性的同时提高了SIFT特征匹配的效率,能够为一些快速性应用提供保障。