Face Verification Using Improved One-dimensional Hidden Markov Model
Abstract
In this paper, we propose an improved version of 1D-HMM for face verification. DCT coefficients of face images are used as observation vectors in HMM states. Three types of have been proposed to improve the overall performance of the classical model: (1) Replacing Baum-Welch algorithm with K.means clustering algorithm, (2) Replacing K-means with adaptive K„means and (3) Adaptive selection of training images the available in data set. The results show identical computational complexity in verification phase and better verification performance compared with other ID-HMM methods. The proposed algorithm has been successfully tested on the ORL face image data set, exhibiting an accuracy of 96%. This is almost 10% higher than the identification rate of the classical ID-HMMs and is comparable with 2D-HMMs accuracy; which basically has much higher processing complexity than 1D-HMM.
Keywords
Face verification, K.means, Discrete Cosine Transform, ORL database, 1D Hidden Markov Model, 2D Hidden Markov Model
