Top-down facilitation of multistage decisions for face recognition

Visual processing in humans is, without a doubt, far superior that that in machines, especially when the end goal is object or face recognition. Neural results from visual object and face recognition in humans provide an excellent model for developing better techniques in machine vision. In this study, we present a particular neural result pertaining to the use of low spatial frequency (LSF) imagery to facilitate recognition of high spatial frequency (HSF) representations of faces and objects and apply it first as a general technique for the classification problem and second as a high-performance recognition method to deal with face recognition on blurry imagery. We demonstrate significant improvement over baseline results using a directly comparable published algorithm. We also discuss the problem and our technique for solving it terms of a mutually beneficial collaboration between the fields of computer vision and neuroscience.