Higronn is a neurodynamical network model explaining mechanisms of perceptual grouping and spatial attention in vision. The model resembles the organization of the visual cortex: a hierarchy of areas, feedforward and recurrent
connectivity and feature-specific tuning. The model carries perceptual grouping in two processing modes:
- When an input image is presented, the model automatically carries rapid feedforward recognition of familiar objects.
- When the focus of spatial attention is directed to a location occupied by a part of an arbitrary object, the model incrementally builds a reliable grouping code for the entire object.
The two modes are modelled with simple and realistic single-unit dynamics that is modulated by attention and not disrupted by variations in input contrast. In sum, the model offers a plausible computational account for contrast-invariant and scale-invariant perceptual grouping.