치아 영상 처리와 신경망을 이용한 신원 확인

Abstract

ABSTRACT

Human Identification Using Dental Image Processing With Neural Networks.

SangJin Park Advisor : Prof. Hyungjun Park, Ph.D. Department of Industrial Engineering Graduate School of Chosun University

The past two decades have seen numerous methods for human identification based on the recognition of bio-mimetic patterns such as iris, gait, face, and finger print. However, all these methods have some trade-offs in terms of cost of equipment, ease of use, recognition rate, and difficulty of falsification. Proposed in this study is a method for human identification using teeth images captured by a PC camera. In the proposed method, the edge profiles of the upper and the lower teeth of each subject are extracted by proper image processing techniques. Then, their geometric patterns are efficiently encoded by neural networks. The teeth images considered in this study are obtained from the frontal face of each subject who opens his or her mouth slightly, maintaining a space between the upper and the lower teeth. As each teeth image contains the subject's mouth and lips, it has always a black-colored region between the upper and the lower teeth, which is not much affected by light conditions. This black-colored region has bio-mimetic information useful for human identification since it contains the edge profiles of the subject’s teeth, which are different from subject to subject. In each teeth image, the back-colored region and its boundary are obtained by boundary extraction techniques using color information. With the minimum enclosing rectangle (MER) of the region boundary, we transform the boundary to make it invariant under rotation. After extracting two edge profiles from the transformed boundary, we approximate them by a pair of polygons from which we compute two difference vectors containing the geometric features of the edge profiles. At last such difference vectors are used as training data of a back-propagation neural network model. In this study, 200 teeth images acquired from 10 university students were processed to obtain the training data of the neural network model. Experimental results have shown that the proposed method has a possible potential that it can be used as an alternative solution for human identification.