基于局部特征约束的压缩感知人脸识别算法研究

Title Compressive Sensing Based on Constraint Local Feature for Face Recognition

人脸识别是计算机模式识别领域中一个研究热点和难点。针对人脸识别中数据量大、高维度、非线性等问题，提出基于局部特征约束的压缩感知人脸识别方法。首先对人脸图像进行选择性约束处理，利用SIFT算法提取人脸图像中的局部特征，以此构成压缩感知算法中的测量矩阵，再利用压缩感知的重构算法计算特征的稀疏表示，在此基础上进行人脸识别。算法在AR人脸库上进行了抗干扰比对测试，实验结果验证了算法对光照、表情以及遮挡等干扰具有强的鲁棒性，局部特征的约束大大降低了特征点的数量，有效提高了人脸识别的正确率。

For the big data, high dimension and nonlinear problems in face recognition, this paper propose a new face recogni-tion method based on compressive sensing with constraint local feature. First process constraint on image, then extract local fea-tures with SIFT method and form a measure matrix, finally we can calculate sparse represent through CS. In this paper, to verify the performance of algorithm do experiments on AR database. Results shows that algorithm can effectively reduce the amount of feature, and have high robustness to illumination, expression and block. Algorithm improves rate of face recognition effectively.