Hardware Architecture for Pseudo-Two-Dimensional Hidden-Markov-Model-Based Face Recognition Systems Employing Laplace Distribution Functions

摘要

A hardware architecture for pseudo-two-dimensional (2D) hidden-Markov-model-based face recognition systems has been developed. The proposed architecture employs the state-parallel organization in which each processing element represents each state in the pseudo-2D hidden Markov model. To reduce the area of processing elements, the mixture of Laplace distributions has been utilized for an observation probability function instead of the mixture of Gaussian distributions. To verify the concept, the proposed architecture has been implemented in a field programmable gate array (FPGA). As a result, the number of logic gates has been reduced by 47% as compared with that using Gaussian distibutions and more than 97% recognition rate has been achieved for the AT & T face database. The processor takes only 44.2 ms for identifying a facial image from 40 people at 100 MHz clock frequency, thus enabling us to build real-time responding systems.