A selective attention model for predicting visual attractors
Éric Dinet, Emmanuel Kubicki · IEEE International Conference on Acoustics Speech and Signal Processing · 2008
The huge amount of visual information continuously received by an observer cannot be wholly analyzed by the brain. In order to interact efficiently with the environment, an observer has to select region of interests in the visual scene. Only the regions of interest will be processed in details by cortical structures. This paper aims at introducing a selective attention model able to predict the location of visual attractors in natural scenes. The underlying idea is to extract and combine, in a competitive process, early visual features such as color and spatial arrangements to construct a saliency map coding interest areas in correlation with human visual behavior. The purpose is to effectively locate which region of a scene would attract the gaze of an observer and then where computational resources should be directed for a selective image processing.