Automatic generation of saliency-based areas of interest for the visualization and analysis of eye-tracking data

Areas of interest (AOIs) are a powerful basis for the analysis and visualization of eye-tracking data. They allow to relate eyetracking metrics to semantic stimulus regions and to perform further statistics. In this work, we propose a novel method for the automated generation of AOIs based on saliency maps. In contrast to existing methods from the state-of-the-art, which generate AOIs based on eye-tracking data, our method generates AOIs based solely on the stimulus saliency, mimicking thus our natural vision. This way, our method is not only independent of the eye-tracking data, but allows to work AOI-based even for complex stimuli, such as abstract art, where proper manual definition of AOIs is not trivial. For evaluation, we cross-validate support vector machine classifiers with the task of separating visual scanpaths of art experts from those of novices. The motivation for this evaluation is to use AOIs as projection functions and to evaluate their robustness on different feature spaces. A good AOI separation should result in different feature sets that enable a fast evaluation with a widely automated work-flow. The proposed method together with the data shown in this paper is available as part of the software EyeTrace [?] http://www.ti.unituebingen.de/Eyetrace.1751.0.html.