A neurodynamical model of visual attention: Feedback enhancement of spatial resolution in a hierarchical system

Human beings have the capacity to recognize objects in natural visual scene s with high efficiency despite the complexity of such scenes, which usually contain multiple objects. One possible mechanism for dealing with this pro blem is selective attention. Psychophysical evidence strongly suggests that selective attention can enhance the spatial resolution in the input region corresponding to the focus of attention. In this work we adopt a computati onal neuroscience perspective to analyze the attentional enhancement of spa tial resolution in the area containing the objects of interest. We extend a nd apply the computational model of Deco and Schurmann (2000), which consis ts of several modules with feedforward and feedback interconnections descri bing the mutual links between different areas of the visual cortex. Each mo dule analyses the visual input with different spatial resolution and can be thought of as a hierarchical predictor at a given level of resolution. Mor eover, each hierarchical predictor has a submodule that consists of a group of neurons performing a biologically based 2D Gabor wavelet transformation at a given resolution level. The attention control decides in which local regions the spatial resolution should be enhanced in a serial fashion. In t his sense, the scene is first analyzed at a coarse resolution level, and th e focus of attention enhances iteratively the resolution at the location of an object until the object is identified. We propose and simulate new psyc hophysical experiments where the effect of the attentional enhancement of s patial resolution can be demonstrated by predicting different reaction time profiles in visual search experiments where the target and distractors are defined at different levels of resolution.