Learning Low-Rank Class-Specific Dictionary and Sparse Intra-Class Variant Dictionary for Face Recognition

摘要: Face recognition is challenging especially when the images from different persons are similar to each other due variations in illumination, expression, and occlusion. If we have sufficient training of person which can span facial that under testing conditions, sparse representation based classification (SRC) achieves very promising results. However, many applications, face often encounters small sample size problem arising number available for person. In this paper, present a novel framework by utilizing low-rank error matrix decomposition, coding techniques (LRSE+SC). Firstly, recovery technique applied decompose per class into matrix. The individual class-specific dictionary it captures discriminative feature individual. represents intra-class variations, such as expression changes. Secondly, combine part (representative basis) supervised integrate all within-individual variant be represent possible between images. Then these two dictionaries used code query image. shared subjects only contribute explain lighting expressions, occlusions image rather than discrimination. At last, reconstruction-based scheme adopted recognition. Since introduced, LRSE+SC handle corrupted data situation not enough samples training. Experimental results show our method state-of-the-art on AR, FERET, FRGC LFW databases.