Shape Identification Using Probabilistic Models of Attentional Sequences

This paper presents a system which demonstrates a biologically motivated approach to artificial vision. As a typical application 2-D shape recognition is chosen. Traditionally recognition problem is divided into two phases: sophisticated preprocessing of the whole image so that regions with similar features can be grouped together and then the recovery of shape parameters of each region using one among the plethora of existing methods. In contrast an approach based on selective attention makes use of simple methods to guide visual processors to the interesting regions of the image. Previous work on selective attention has focused on pre-attentive and attentive stages while almost no work has been done on the final stage cognition. This paper presents the beginnings of how to complete the pre-attention, attention, cognition sequel of selective attention. An approach to shape recognition based on probabilistic models of attentional sequences is described and the capabilities of the system are investigated. Experiments performed on rotated shapes demonstrate the performance of the algorithm.