In this paper a novel Mahalanobis Distance based method for sparse representation classification was designed to improve the recognition efficiency for different illumination condition face images. Mahalanobis Distance and Cholesky decomposition are introduced to solve the sparse solution vector, and Mahalanobis Distance based Sparse Representation Classification (MSRC) is designed to recognize the face image. Firstly, Mahalanobis Distance based L1 -minimization algorithm is proposed to obtain the sparse representation. Then, reconstruct the test image. Finally, the one that has the minimum reconstruction error is selected as the most matched face. Compared to the traditional SRC algorithms, our algorithm significantly reduces the influence of illumination. Lots of numerical experiments based on ORL face database and Extended Yale face database B are performed. The results show that the proposed Mahalanobis Distance based Sparse Representation Classification algorithm can achieve about 97% recognition rate for normal face images.%常用分类算法对人脸图像在不同光照条件下的识别效果较不理想.设计了一种新颖的基于马氏距离(Mahalanobis Distance)的人脸识别分类算法(Mahalanobis Distance based Sparse Representation Classification,MSRC).该算法框架基于稀疏表示原理,通过引入马氏距离和乔里斯基分解( Cholesky decomposition)求出最优稀疏解向量,最终实现人脸特征分类识别.算法首先求解基于马氏距离的最小L1范数,进而对测试样本实现稀疏重构,并通过判断重构样本与原始样本的残差值最终完成分类.与传统稀疏表示分类算法相比,该算法显著降低了光照对人脸图像的影响.在Extended Yale face databaseB人脸库上的实验结果表明,所提出的基于马氏距离的稀疏表示分类算法能达到97％的分类效率,并且在人脸不同光照情况下仍能得到较好的识别效果.
展开▼
