摘要： 针对人脸识别中特征提取和特征分类问题，提出一种基于SIFT特征和Fisher鉴别的人脸识别新方法。采用具有旋转、缩放、平移、光照不变性及部分仿射不变性的SIFT特征作为初级特征，利用Fisher线性鉴别方法再次提取初级特征，从而得到样本的二次特征，通过比较二次特征之间的欧氏距离，得到识别结果。实验结果表明，新的方法具有99.65%的正确识别率，高于Fisher方法和核Fisher方法，识别速度和Fisher方法相当。

关键词: SIFT特征, Fisher线性鉴别, 错误接受率, Adaboost算法

Abstract: Aiming at the problem of feature extracting and feature discriminate in face recognition, this paper introduces a new method of face recognition based on Scale Invariant Feature Transform(SIFT) feature and Fisher discriminant. The SIFT, with the invariant of rotation, scale, translation and illumination, can be used as a primer feature, then it uses FLD to extract the quadratic feature, by comparing the Euripides distance of the quadratic feature. It can find right class of the test picture. Practice illustrates that the recognition rate of new method is 99.65%, which exceeds FLD and KFLD, and the recognition speed of new method is corresponding to Fisher methed.

Key words: Scale Invariant Feature Transform(SIFT) feature, Fisher linear discriminant, False Acceptance Rate(FAR), Adaboost algorithm

中图分类号: 

• TP391