Ear recognition using bilinear Probabilistic Principal Component analysis and sparse classifier

Biometric systems are becoming more and more popular with the increase in need for strong security systems. Ear is one of the biometric trait whose structure cannot change in the course of human life. In this paper ear recognition based on two dimensional Probabilistic Principal Component analysis (PPCA) using sparse representation algorithm classifier is proposed. The main objectives of the work are to validate (i) Ear recognition gives good recognition performance in two dimensional PPCA methods compared to non-probabilistic methods; and ii) Sparse representation classifier gives good performance as compared to Euclidean distance classifier. Initially ear images are normalized using techniques such as Retinex and CLAHE and the ear region is segmented by fitting best fit ellipse on the edge detected image. Bilinear PPCA is employed as feature extraction in which the principal axis is estimated efficiently from the covariance matrix by maximizing the estimates of log-likelihood function. Classification is done based on the obtained features using Euclidean distance and sparse representation algorithm classifier. The performance of the proposed algorithm is evaluated using USTB Ear database and OWN database. Results show that, probabilistic models give a good performance compared to non-probabilistic models thus BPPCA outperforms 2DPCA and GLRAM. Sparse representation algorithm classifier shows good performance compared to distance classifier.