Application of hopfield neural network for facial image recognition

Abstract

Hopfield model (HM) classified under the category of recurrent networks has been used for pattern retrieval and solving optimization problems. This network acts like a CAM (content addressable memory); it is capable of recalling a pattern from the stored memory even if it’s noisy or partial form is given to the model. This paper presents a framework to use HM for machine recognition of human faces. The approach presented here uses Otsu’s method to transform facial images (low and ultra-low-resolution) from grey scale to binary, and Hebb rule to store them in the W (weight matrix) of the network. HM is then tested with up to 45% distortion in facial images; the network is allowed to evolve asynchronously to a stable state. To check positive retrieval, we match (bit-by-bit) the stable state facial image with the original facial image. Our results show 100% retrieval for grey scale facial images (of 60×60 pixel size) for up to 30% distortion. This suggests that HM can be used for face-based security applications when the number of individuals allowed is a limited number like a high-security military area or a lab.