A model of the visual attention to speed up image analysis

Current methods in image analysis are rather inefficient. They are too systematic and application specific. Parallel processing and learning procedures help to make up for those drawbacks. The intrinsic complexity of the visual recognition however prevents one to complete the analysis of a scene in such a parallel way. We need a smart system to guide the analysis of small regions of interest. Psychologists have suggested that it could be the role played by the attention system in the human brain. Their models are based on the extraction of basic visual features and variable magnifications. They select first isolated singular and expected objects. The model we suggest in this article is an implementation based on artificial neural networks that integrates knowledge of the physiology of the brain. We use the local geometry to build our basic feature maps. Simulation results are very promising