Two-Phase Test Sample Representation with Efficient M-Nearest Neighbor Selection in Face Recognition

Sparse representation method, especially the Two-Phase Test Sample Representation (TPTSR) method is regarded as a powerful algorithm for face recognition. The TPTSR method is a two-phase process in which finds out the M nearest neighbors to the testing sample in the first phase, and classifies the testing sample into the class with the most representative linear combination in the second phase. However, this method is limited by the overwhelming computational load, especially for a large training set and big number of classes. This paper studies different nearest neighbor selection approaches for the first phase of TPTSR in order to reduce the computational expenses of face recognition. Experimental results and theoretical analysis show that computational efficiency can be significantly increased by using relatively more straightforward criterions while maintaining a comparable classification performance with the original TPTSR method.