Research of increasing virtual face samples for small sample problems and its applications in face recognition

In order to solve the small sample problems and the linear inseparable problems caused by some nonlinear factors, this paper proposed a method to generate multiple virtual samples similar to the original images by its class, then all virtual samples were combined as a new database for training. The method not only helps to increase more samples, but strengthens the reliance of virtual samples on the samples in original database. Since the face images are high dimensional, principal component analysis (PCA) is used for dimension reduction and feature extraction. The experiments based on the ORL face database show that the recognition rates have been greatly improved and the recognition results are relatively stable with the increased sample method.