Skip to Main Content
The sensorimotor system of saccadic scene analysis is investigated in a multi-disciplinary research project. Eye-movements of human observers are recorded for a variety of scenes. Our statistical evaluations indicate that fixations are selected on the basis of higher-order statistical properties. The structure of the polyspectra suggests that nonlinear operators with a selectivity for intrinsically two-dimensional image features serve as “saliency detectors”. The bottom-up feature extraction is combined with a knowledge-based system into an integrated architecture. in which a saccadic fixation corresponds to the “best question to ask” given the internal state and the recent experience of the system.