Face Recognition Using Gabor-Based Feature Extraction and Feature Space Transformation Fusion Method for Single Image per Person Problem

摘要

Discriminant analysis technique plays an important role in face recognition because it can extract discriminative features to classify different persons. However, most existing discriminant analysis methods fail to work for single image per person problem (SIPPP) because there is only a single training sample per person such that the within-class variation of this person cannot be calculated in such case. In this paper, we present a new face recognition method for SIPPP. The method is a combination of Gabor wavelets, feature space transformation (FST) based on fusion feature matrix, and nearest neighbor classifier (NNc). First, we use Gabor wavelets to extract the feature vectors from a raw training sample image, because Gabor-based features are more robust than spectral-based features and could avoid the local distortions caused by the variance of expression, pose, light and noise. Then, the extracted spatial-based feature vectors and spectral-based feature vectors are combined, and projected to a low-dimensional subspace by using dimensionality reduction techniques. Finally, the classification can be completed via using NNc. The proposed method is abbreviated as G-FST. The performance of G-FST method is evaluated on ORL, Yale and FERET databases. Experimental results show that the G-FST method outperforms the other related methods in terms of recognition rates and recognition time.