Object Selection with Dynamic Neural Maps

This contribution presents an approach for object selection. It is based on the functional role of attention in its control of action, that is the selection in space and the selection at the object level to serve the extraction of further object related features. The selection is performed on topographic maps which describe by their local activation the existence of an object hypothesis and is achieved by a dynamical two step process of i) competition and ii) region aggregation. The neural dynamics is based on analog neurons, mathematically described with a nonlinear activation dynamic. This method is appropriate for all objects which are detected by their local features, like color or texture. The performance is demonstrated on two real world selection problems.