A biologically plausible approach to cat and dog discrimination

Abstract

The paper describes a computational model of human expert object recognition in terms of pattern recognition algorithms. In particular, we model the process by which people quickly recognize familiar objects seen from familiar viewpoints at both the instance and category level. We propose a sequence of unsupervised pattern recognition algorithms that is consistent with all known biological data. It combines the standard Gabor-filter model of early vision with a novel cluster-based local linear projection model of expert object recognition in the ventral visual stream. This model is shown to be better than standard algorithms at distinguishing between cats and dogs.