Abstract:Robustness to noise, outliers, and corruption is an important issue in image feature extraction. A discriminative low-rank embedded image feature extraction algorithm is proposed in this paper to address this problem. Firstly, manifold con-straints are introduced based on the low-rank embedding (LRE) approach to capture the geometric structure of the local manifold, taking into account both local and global information. Secondly, a discriminant analysis term is also introduced to obtain global discriminant information and learn the optimal projection matrix for data dimensionality reduction. Finally, the test samples are projected into a low-dimensional space for classification. Numerical experiments show that the classi-fication accuracy of the method proposed in this paper is 95.62%, 95.22%, 86.38%, and 86.54% on the ORL, CMUPIE, AR, and COIL20 datasets, respectively, which is better than that of the dimensionality reduction-based image feature ex-traction algorithm.