判别共同法向量方法结合格拉斯曼流形的人脸识别

Face Recognition Based on DCVA and Grassmann Manifold

提出了判别共同法向量结合格拉斯曼流形的人脸识别方法。使用自熵图像表示人脸;利用判别共同法向量方法提取特征,并利用字典学习优化特征矩阵;将提取的特征表示在格拉斯曼流形上,对格拉斯曼流形上的点进行有效聚类,并利用投票方法完成分类。在Caltech和AR人脸数据库上的实验结果表明,相比其他几种较新的表情和光照人脸识别方法,提出的方法取得了最高的识别率,且明显减少了执行时间,对人脸表情和光照变化具有鲁棒性。

Due to the influence of expression and illumination changes on the recognition performance, face recognition becomes difficult. This paper proposed the discriminant common normal vector combined Glassmann manifold method for face recogni- tion. First of all, this method usied self quotient image to express face; then, discriminant common normal vector method was used to extract features, and the dictionary learning was used to optimize feature matrix; finally, the extracted features were projected to Glassmann manifold, the points on Glassmann manifold were clustered, and classification was completed by using the voting method. The experimental results from Caltech and AR face database, show that compared with several other rela- tively new methods of face recognition, the proposed method has better recognition performance, and has robustness to facial expressions and illumination changes.