The Face Recognition Algorithm Based On Principal Component Dimensionality Reduction

As a typical research topic in pattern recognition and image processing,face recognition is not only important in theory,but also has important application prospect in security and finance.Although the face recognition technology has been promoted in the business,but how to improve recognition accuracy in case of interference of light variation and facial expression has been focus of the field of research and difficulties.This paper focuses on the difficult situation of face recognition,and studies the shortcomings of the traditional face recognition method,which is sensitive to exposure and expression of face images and has large intra-class dispersion.The main contribution of our work is as follows:First of all,based on summary of traditional face recognition technology,this paper focuses on Gabor wavelet filtering algorithm and the face recognition algorithm based on principal component analysis,we also introduce the main ideas and implementation of these two algorithms.Then,for the implementation of traditional 2DGabor and 2DPCA,there are some cases in which information is lost and there is a large intra-class correlation.This paper proposes an algorithm to replace the 2DPCA by using Cross-covariance Two-dimensional Principal Component Analysis(Cro2DPCA).This algorithm uses crossover operation method to calculate the left and right mapping matrices,and then uses the nearest neighbor classifier to classify and obtain the recognition result.Finally,based on the algorithm proposed in this paper,a face detection and recognition system is built up with MATLAB.The system can capture image from camera,detect face image,preprocess the face image,extract features,and then output recognition result.Experimental results of the face recognition algorithm combined with 2DGabor and Cro2 DPCA in the ORL database,AR database and FERET database show that the proposed method is superior to traditional face recognition algorithm and its improved algorithm.Whether in small training samples or large training samples in the case can take into account the degree of simplicity and accuracy,effectively improve the recognition performance.