摘要： 针对光照、表情、遮挡、几何变化等情况下的人脸识别问题，提出一种基于尺度不变特征变换(SIFT)和增强Hough变换的人脸识别方法。利用SIFT的位置、尺度、方向和描述符4个信息，通过增强Hough变换消除错配，将候选匹配点和离散点距离进行加权累计获得高辨别力的匹配分。在AR人脸数据库上的测试结果表明，在只选取每人1幅图像作为模板的情况下，能够获得近99.6%的识别率。

关键词: 尺度不变特征变换, Hough变换, 人脸识别, 特征提取, 匹配策略

Abstract: Aiming at the difficulties of face recognition such as change in illumination, expression variation, partial occlusion and geometrical transformation, a face recognition method based on Scale-invariant Feature Transform(SIFT) and enhanced Hough transform is proposed. By using four parts of SIFT features and the enhanced Hough transform, the false matches are removed effectively, and the score having high discrimination ability is obtained through weighting more information. Experimental results performed on the AR face database demonstrate that about 99.6% of recognition rates are achieved under selecting only one image per person as template.

Key words: Scale-invariant Feature Transform(SIFT), Hough transform, face recognition, feature extraction, matching strategy

中图分类号: 

• TP391