Face Recognition by Observation-Sequence-Based Methods Based on Pseudo 2D HMM and Neural Networks

Face recognition is an obviously interesting research area, due to its applicability in a biometric system both in commercial both in security fields. In this paper a Pseudo 2-Dimension Hidden Markov Model (P2D-HMM) combined with three different observation-sequence-based methods is introduced for face recognition. The P2D-HMM proposed, is applied to five RoI (Region of Interest) of images, one for each significant facial area in which the input frontal images are sequenced: forehead, eyes, nose mouth and chin. It has been trained by coefficients of an Artificial Neural Network used to compress a bitmap image in order to represent it with a reduced number of significant coefficients manipulated by the three observation-sequence-based methods. The introduced system, applied to the input set consisting of the Olivetti Research Lab. face database integrated with others photos, allows to obtain an high rate of recognition, up to 100% in particular with the P2D-HMM realised by the 'Strip'-like sequencing method.