Hopfield like networks for pattern recognition with application to face recognition

This work will discuss the use of a robust Hopfield like structure for pattern recognition and face recognition. The design of the energy function and the mapping of the recognition problem to the realm of graph matching have led to the robustness of the algorithm detailed here. Examples of the success obtained from this network using various local feature spaces will be examined. This will include examples associated with recognition in a multicontext environment and face recognition under pose rotation in depth. The Hopfield neural network (HNN) utilized in this work is the annealed Hopfield neural network (AHNN). The marriage of the HNN and mean field annealing (MFA) has led to a robust near optimal algorithm for combinatorial optimization. We will discuss the formulation of a recognition algorithm based on the optimizing energy function of the Hopfield neural network and thus the AHNN We will also present the use of the AHNN for face recognition (FR) by developing a novel feature space termed anthropometric face features.