Performance evaluation of face recognition algorithms

Abstract

Face recognition consists of computerized recognition of personal identity based on statistical or geometric features obtained from face images. Many face recognition techniques have been developed; among them are Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and Bayesian Intrapersonal/Extrapersonal Classifier (BIC). The BANCA database is a multi-modal database developed for training and testing purposes. In this thesis software tools were used to carry out comparative analysis of PCA, LDA and BIC using the BANCA database. These tools retrieve and preprocess images (cropping, grayscaling, normalizing and histogram equalization) from sequential records within the BANCA database for algorithm evaluation. The PCA, LDA and BIC algorithms were then integrated using the Colorado State University Face Identification System (CSUFIS). Finally a verification environment over the set of images to be tested was developed, the above algorithms were invoked over the verification set, and verification parameters were collected. Results proved PCA performed most accurately and efficiently with an average recognition rate of 83.8%, while BIC lagged behind with an average recognition rate of about 72%. Although BIC was the least efficient in terms of speed, LDA was the least accurate in terms of identification, as it scored an average recognition rate ranging from 72% to 65%.