ABSTRACT
Our goal is to understand the role of natural signal processing by simulating it on machine perception tasks. We use Gabor filters as a first approximation to the operation of primate striate cortex. It has been proposed that besides modeling the receptive fields of simple cells in striate cortex (Jones & Palmer, 1987), Gabor receptive fields provide an efficient and sparse code for natural images (Field, 1994; Daugman, 1989). It has also been proposed that one of the functions of this sparse code may be to facilitate visual pattern recognition (Field, 1994). We test this prediction on a machine vision task: automatic, speaker independent lipreading of the first four digits in English.
