改进的SURF特征提取与匹配算法

The improved algorithm for SURF feature extraction and matching

针对SURF图像匹配算法在求取特征点主方向阶段太过依赖局部区域像素梯度方向,造成找到的主方向不准确、误匹配率高的问题,提出一种改进的SURF匹配算法。首先,利用SURF提取出特征点的64维特征向量并构建描述子;然后,采用距离匹配测度和余弦相似度匹配测度相结合的方法进行特征点匹配;最后,使用改进的随机抽样一致算法,进一步提高匹配正确率。实验结果表明:图像在一定程度的旋转、缩放、模糊、光照和视角变化情况下,改进算法匹配正确率在94%以上。

Image matching algorithm based on SURF relies on gradient direction of local area pixels too much in the stage of getting a major orientation to each feature point,which causes the inaccuracy of major orientations and high false matching rates. Aiming at above problems,the paper introduces an improved algorithm for SURF matching. Using 64-dimensional feature vectors of feature points extracted from SURF,it builds the descriptors,and applies the method of combining metric of distance matching with metric of cosine similarity matching to match feature points,makes the improved RANSAC adopt to further lower the rate of false matching. The experimental results show that under the conditions of images rotating,zooming,blurring lighting and view change,the accuracy of the improved algorithm can reach over 94%.