Professor, Computer Science,
Director, Centre for Vision Research, York University

 

Abstract
A new model of attentive visual motion processing is presented. This is a feedforward motion-processing pyramid whose motivation lies in the neurobiology of primate motion processes. On this structure the Selective Tuning Model (STM) for visual attention is implemented and demonstrated, showing how it can localize and label simple motion patterns. The motivating goal for the work was to demonstrate that the STM could operate as originally described on a biologically realistic feature pyramid. This has been successfully shown; however, the motion pyramid itself has also emerged as an important contribution. The motion decomposition is the contribution, and not so much its form, this being the first model to explicitly include local spatial derivatives of velocity as an intermediate representation.

 

As a prelude to the description, the STM is briefly overviewed with a focus on the overall algorithm and winner-take-all competitions that are the basis of the selection process. As part of this, a new solution to aspects of the binding problem is proposed and shown to be sufficient for the task of grouping motion features into coherent object motion. Several examples of STM performance on image sequences are shown.