Effectiveness analysis of selected attention models

The aim of this study was an empirical verification of five different computer models of exogenous human attention. These models allow to predict the likely fixations on a static image and calculate saliency maps which show how distinctive the image elements are. The basis for the verification of the models was a modified cuing task , originally developed by Posner. Using the algorithms provided by the authors of the models an analysis of selected images was performed. Then the effectiveness of the models was verified in experimental studies: an image was presented to the subject, then on this image a visual stimulus appeared and the reaction time was measured. It was assumed that the emergence of a stimulus at the point where previously an attention attracting element appeared would accelerate the reaction. The study was conducted on a set of over 300 different, emotionally neutral, natural images. Obtained results indicate the potential usefulness of the proposed schema for testing the efficiency and refinement of attention models.