Study Of Single Sample Problem Of Face Recognition Based On Gabor Wavelet

Face recognition is an important subject in artificial intelligence field, and it has gained extensive attention from researchers in the past decades. Due to the good characteristic of human face image's Gabor wavelet feature, face recognition technology based on Gabor wavelet is a very popular method. Face recognition is an intersecting subject, and this paper is focusing on feature extraction of face recognition, and also this paper proposes a new solution for single sample problem in face recognition.This paper utilizes the property that a complex number can be resolved into magnitude and argument in a polar coordinate system, extracts argument features from human face image's Gabor wavelet representation, and gives the distribution figure of human face image's Gabor wavelet representation containing argument features and magnitude features.For the single sample problem, by making use of magnitude features and new extracted argument features, this paper proposes a novel Enriched Gabor feature based Principal Component Analysis (EGPCA) algorithm. Based on the EGPCA algorithm, this paper implements a face recognition system, and compares EGPCA with ( PC )2A, E (PC)2A, and SVD Perturbation in a face recognition task. Experimental results on FERET face database show that EGPCA can achieve 89.5% recognition rate when only one training image per person is available in face recognition, which is superior to other three algorithms.For the purpose of comparing the efficiency of argument features and phase features in face recognition, this thesis applies these two kinds of features in face recognition experiments based on FERET and ORL face databases. Experimental results on both two face databases show that argument features are superior to phase features.