Efficient Face Recognition Based On Implicit Gabor Features In Frequency Domain

Face recognition is a kind of biometric identification technologies. It analyzes face images by computer to extract effective information for identification, which has an extensive application in public security investigation, credit card identification and monitoring system. Face recognition is one of the most challenging issues in pattern recognition and computer vision.This thesis reviews the background, significance, development process and current situation of research on face recognition, then classifies and introduces the typical methods of face recognition. Through analyzing face recognition algorithm based on Gabor wavelet features and PCA dimension reduction, we discover that the method is inefficient due to the calculation on multi-scale and multi-angle Gabor wavelet features. To solve this problem, an improved principal component analysis algorithm using Gabor features rather than directly calculating the features is proposed. First, a filtering weight matrix independent of input images can be offline calculated. The information of Gabor features is hidden in the matrix. After getting the PCA covariance matrix, we further implicitly extract the Gabor features of the images,and perform dimension reduction on them according to Parseval's relation in two-dimensional frequency domain. Finally, a linear support vector machine is employed for classification and recognition.Compared with traditional method, the proposed implicit Gabor-PCA algorithm in frequency domain shows a better performance. It significantly reduces the dimension of features, decreases the computation and improves the efficiency of identification. Experiments on FERET, ORL, YALE face databases show that under the same condition, the recognition speed of proposed algorithm has a large enhancement and the recognition accuracy rate is also slightly higher.