Scene analysis with saccadic eye movements: Top-down and bottom-up modeling

The perception of an image by a human observer is usually modeled as a para llel process in which all parts of the image are treated more or less equiv alently, but in reality the analysis of scenes is a highly selective proced ure, in which only a small subset of image locations is processed by the pr ecise and efficient neural machinery of foveal vision. To understand the pr inciples behind this selection of the "informative" regions of images, we h ave developed a hybrid system that consists of a combination of a knowledge -based reasoning system with a low-level preprocessing by linear and nonlin ear neural operators. This hybrid system is intended as a first step toward s a complete model of the sensorimotor system of saccadic scene analysis. I n the analysis of a scene, the system calculates in each step which eye mov ement has to be made to reach a maximum of information about the scene. The possible information gain is calculated by means of a parallel strategy wh ich is suitable for adaptive reasoning. The output of the system is a fixat ion sequence, and finally, a hypothesis about the scene. (C) 2001 SPIE and IS&T.