Abstract: It is difficult in the image recognition technology with different rotation angles and the change of image matching, and SURF algorithm in multi-angle feature matching process has more noise, easy to mismatching, and matching efficiency is low. Combination with cluster and Mahalanobis distance, this paper proposes an improved multi-angle SURF image matching algorithm. First, it uses clustering algorithm to eliminate the noise, to the feature point SURF algorithm extracted, uses clustering algorithm to classify and remove noise to get the new feature point data set. Then, it uses Mahalanobis distance’s characteristics that it considers the overall correlation, and has the characteristics of affine invariance, replacing Euclidean distance with Mahalanobis distance. When the experiment is applied to multi-angle image matching, compared with the original SURF, the improved algorithm has obviously improved on the matching efficiency.

Key words: image matching, Speeded Up Robust Features（SURF） algorithm, cluster algorithm, Mahalanobis distance, affine invariant

摘要： 对具有不同旋转角度和变化的图像进行匹配是图像识别中的技术难点，SURF算法在多角度图像的特征点检测和匹配过程中存在易受噪声点干扰、产生误匹配从而导致匹配效率低等不足。结合聚类和马氏距离，提出一种改进的多角度SURF图像匹配算法。首先利用聚类算法对原有算法提取的特征点进行噪声剔除处理，生成新的特征点数据集；然后利用马氏距离能够有效考虑整体相关性及其具有仿射不变性等特点，将SURF算法中的欧式距离用马氏距离替代。实验应用于多角度图像匹配时，改进算法较原SURF算法在匹配效率和准确率上有明显提高。

关键词: 图像匹配, 快速鲁棒特征（SURF）算法, 聚类算法, 马氏距离, 仿射不变性