A scene registration method based on a dynamical receptive field model of biological vision

摘要

A scene registration method based on a model of dynamical receptive field organization of biological vision is presented. In this model, the receptive fields of simple cells have differently oriented Gabor-type receptive field functions. Collectively, the simple cells in a hyper column extract an HC-vector from local sensory input serving as the place token for the scene registration. By incorporating a data driven dynamical tuning mechanism of simple cells, the place tokens are invariant to distortions. The initial experiments show that this biologically motivated method is accurate and robust.