A pre-attentive model of biological vision

Most view-based models have neglected two important facts about object learning: the transformation from retinal to cortical representation is space-variant, and the eyes move actively to explore the world. The high resolution foveal representation of retinal image facilitates object recognition, and the low resolution peripheral representation provides suitable commands for eye movements. In this paper, pre-attentive processing mechanism of biological vision is studied, including cortical magnification, contrast enhancement, boundary processing and surface filling mechanisms. Compared with traditional biological visual image processing system, the proposed pre-attentive model is more biologically plausible, and simulation results indicate that it can provide input to higher-level visual cognition process as expected.