A multiacuity connectionist model for local speed estimation

Abstract

Multiresolution foveal imaging offers a field-of-view × acuity × frame rate product much greater than that of uniform acuity imaging, and can reduce the computational complexity of vision in dynamic scenarios with localized relevance. The latter requires space variant retinotopic processing with results that are consistent between different resolutions. A physiologically inspired connectionist model is presented which yields local speed estimates consistent between coarse resolution peripheral vision and fine resolution central vision. A non-uniform acuity profile across the field-of-view is compensated by an inverse profile of weights in the model. This model demonstrates the efficiency of multiresolution vision when processing and retinotopology are properly matched.