A simplified second-order HMM with application to face recognition

In this paper, we propose a novel approach to simplify the second-order 2-D HMM as applied to the problem of Face Recognition (FR). The proposed approach exploits the nonoverlapped feature block conditions and the independence that arises in the conditional statistical relationship between feature blocks in close neighborhoods. System performance is studied and the impact of the number of states and the kernels of the state probability density function is highlighted. The system was tested on the facial database of AT&T Laboratories Cambridge [1] and a recognition rate up to 100% has been achieved with relatively low complexity.