Single Sample Face Recognition: Discriminant Scaled Space vs Sparse Representation-Based Classification

摘要

Sparse Representation-based Classification (SRC) is an effective solution of face recognition as there have been many studies around it. However, classical SRC needs a large train data for the galley to produce an over-complete dictionary which result in high accuracy. This paper purposes to show that when there is only one sample per subject for the gallery, the simple linear Discriminant Scaled Space (DSS) can outperform classical SRC and is competitive with new single sample version of that along with significantly less runtime. In addition, it will be shown that SRC methods can be computed on the data projected to DSS which result in higher accuracy with less run time. To show the effectiveness of DSS, it is compared with different kinds of SRC on 11 public databases.