Performance Evaluation of Principal Component Analysis and Linear Discriminant Analysis for Human Teeth Recognition

Abstract

Biometric identification methods have been proved to be very efficient, natural, and easier for users than traditional methods of human identification. Biometric is defined as the science of recognizing a person based on certain physiological (fingerprints, face and voice) traits which possess low discriminating contents; these change over time for each individual. Thus, these biometrics show lower performance as compared to the strong biometrics (eg. fingerprints, iris, retina, etc.). Among various physiological biometrics, teeth biometrics has been found to be interesting and promising in the biometrics field. In this thesis, for the performance evaluation of appearance-based statistical methods, both Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are tested and compared for the recognition of human teeth images. In the transformed space, euclidean distance classifier is employed. Teeth were acquired using a simple low-cost setup consisting of a digital camera. Three sets of experiments are conducted for relative performance evaluations. In the first set of experiments, the recognition performances of PCA and LDA are demonstrated. The effect of illumination variations is evaluated in the second set, whereas teeth images are anterior and posterior occlusion in the third set of experiments. The goal of this thesis is to present an independent and comparative study of two most popular appearance-based teeth recognition algorithms (PCA and LDA) in various conditions.