SIFT特征匹配算法优化

Optimization of SIFT Feature Matching Algorithm

基于SIFT特征的图像目标匹配与检测应用中,特征点的误匹配将直接影响系统对目标检测的灵敏度。提出了一种将二分查找法与仿射变换结合,用于目标快速检测的方法。先用算法复杂度较低的最近邻-次近邻法对SIFT特征点进行粗匹配,并用匹配对中两特征点与主方向角度差进行筛选。随机选取若干组特征点,计算其仿射变换参数,依据其参数的统计分布特点,用二分查找法得到最优解,实现对目标的检测。结果显示,算法有较高的检测效率和稳定性。

Based on SIFt feature matching and image target detection applications, mis- matching feature points on the target system will directly affect detection sensitivity. A bina- ry search and affine transformation combined method is presented for rapidly detection of the target. Firstly, rough match is performed with a lower computational complexity nearest neighbor - next nearest neighbor method of SIFT feature points. Then, screening has been done with two matching feature points in main direction of the angular difference. Based on the statistical distribution of its parameters, some set of feature points are selected randomly to calculate the affine transformation parameters within the binary search method, by which the optimal solution is obtained and target detection is realized. The results show that the al- gorithm has higher detection efficiency and stability.