Teeth selection using particle swarm optimization

Abstract

Usually, the forensic human identification through dental biometrics is performed manually by human experts. An important alternative to facilitate and expedite this task is to use an ADIS (Automated Dental Identification System). In ADIS, the selection of most discrimintive teeth can improve substancialy the subject identification rates. In this paper we propose the use of BPSO (Binary Particle Swarm Optimization) in order to select optimum subsets of most discriminative teeth shape descriptors, and, consequently, to improve the rates of correct person identification. Results obtained in our experiments have confirmed our hypothesis that some teeth are more discriminative than others and should be priorized during the forensic human identification based on dental recognition. The use of BPSO have reduced the EER from 20% to 10%, and have reduced the rank-k from k=15 to k=4 in order to reach 100% of correct teeth retrieval.