A nearest neighbor classifier based on virtual test samples for face recognition

Abstract

In this paper we propose a nearest neighbor classifier which aims at improving the classification accuracy of face recognition. The idea of the proposed method is as follows. Firstly, the symmetry of the original test samples is used to generate new test samples. Then, training samples are used to represent original test samples and the virtual test samples respectively. It takes the advantage of the weighted sum to construct a nearest neighbor classifier to improve the accuracy of face recognition. Meanwhile, the proposed method codes a test sample as a linear combination of all of the training samples, and the deviation between the training samples and the test samples is exploited to classify the test sample. The proposed method can perform better in the case with a small number of training samples than the improvement nearest neighbor classifier. In this paper, the proposed method is compared with a simple and fast representation-based face recognition method, an improvement to the nearest neighbor classifier, a novel sparse representation method based on virtual samples for face recognition (SRMVS) and a two-phase test sample sparse representation method (TPTSR). The experimental results show that our method has better classification results than the others.