Research On Face Recognition Algorithm Based On Principal Component Analysis

Face recognition technology is an important technology of biometric identification and aims to extract the features of face images for identity verification.In addition,face recognition is not only the difficulty of pattern recognition community,but also the key part of numerous disciplines,e.g.signal processing,intelligent control,etc.More specifically,face recognition mainly consists of three parts:face image preprocessing,feature extraction and classifier design.It is notable that research on these three parts is meaningful to develop practical face recognition system eventually.In this dissertation,we will investigate principal component analysis(Sub-pattern PCA),which is a widely-used feature extraction technology in face recognition.Motivated by two dimensional PCA,we extend two PCA technologies based on sub-pattern,e.g.sub-pattern principal component analysis(Sub-pattern PCA,Sp-PCA)and progressive principal component analysis(Progressive PCA,Pr-PCA)into a 2D-version.Then,we also study a face image preprocessing technology based on weighted wavelet transform and propose a new weights assignment scheme.The detailed contributions are as following:(1)Firstly,this dissertation deeply studies the conventional face recognition technology and relevant improved algorithms.Then,based on the popular standard face databases,we simulate,compare and analyze these algorithms(2)Secondly,motivated by 2D-PCA,we propose the 2D extension of Sp-PCA and Pr-PCA,which are denoted by 2D-SpPCA and 2D-PrPCA for feature extraction of face recognition.These two 2D sub-pattern based PCA algorithms inherits the superiority of 2D-PCA and owns low computation.Meanwhile,they are also able to capture the local information of face images as well as sub-pattern based PCA.Consequently,2D-SpPCA and 2D-PrPCA will be more suitable for coping with feature extraction of face images under different environments such as lighting condition,expression and pose variation.Finally,we conduct extensive experiments to evaluate the performance of our proposed 2D-SpPCA and 2D-PrPCA.(3)Finally,in order to eliminate the interference of illumination,expression and posture etc.to face images,this dissertation presents a face image reconstruction method based on weighted wavelet transform.In this method,we firstly determine a suitable wavelet basis function and decomposition level based on a series of experiments.Then,we describe a novel weights assignment scheme of wavelet components to reconstruct the face images for further feature extraction.Finally,we combine the weights assignment scheme with 1D sub-pattern based PCA and 2D sub-pattern based PCA and propose two face recognition methods.Experimental results on the numerous face databases demonstrate that our weights assignment scheme achieves a satisfied improvement compared with several state-of-the-art schemes.At the same time,the improved algorithms are presented based on weighted wavelet transform,which are 1D sub-pattern based PCA combined with it and 2D sub-pattern PCA combined with it,experimental results are superior to other feature extraction methods.