Recognition of human faces in distorted images based on principal component analysis and gabor wavelets

Abstract

The technology of human face recognition is now mature enough to be widely used in security by law enforcement agencies for dealing with criminal activities, government agencies for border control as well as government organizations and private enterprises for access control. However, there are still a lot of room for improvement. In this thesis, we study strategies to improve the recognition accuracy. Two proposed schemes, Principal Component Analysis (PCA) and Gabor wavelet, for human face recognition are discussed and the implementations of sub-modules in each scheme are introduced. In principal component analysis based human face recognition, we started from the basic method by first identified and developed the major testing criteria. Once the major testing criteria were developed, we analyzed those criteria which had significant impacts on the accuracy of recognition. More face databases were used in order to positively identify those criteria. Distorted images with poor illumination at the background had been identified as one of the major impacts on the accuracy of recognition. We then further investigated on this criteria and found ways to improve the results of recognition. In Gabor wavelet based human face recognition, we first examined the classification capability of 40 different basic Gabor phase representations. We utilized those Gabor features from facial images, tested on the selected distorted images and compared the results and findings with the principle component analysis based human face recognition.