Virtual Training Samples and CRC based Test Sample Reconstruction and Face Recognition Experiments

Abstract

Sparse Representation (SR) has the merit to associate each test sample properly with the training samples. Collaborative representation classification (CRC) is a well-known generalized SR method and has achieved outstanding performance in Face Recognition (FR). In this paper, we propose an improvement to CRC, which combines the original training sample and mirror virtual face to form a new training set, uses this new training set to rebuild the test sample and then performs a two-step classification. The face recognition experiments show that the proposed method outperforms CRC and has certain robustness.