Image Segmentation Scrutiny by Eye Tracking

Abstract

Among the visual functionalities the most popular investigations refer to free viewing scan-paths, controlled fixation in visual search in the form of task-driven (top-down) visual guidance, priming mechanisms i.e. The procedural changes when target and distract or features are repeated on consecutive trials, and stimulus-driven (bottom-up) analysis mainly of simple artificial images, with discrete search items features, but rarely on ordinary real life scenes. Here we are pointing out a different task derived directly from the image processing, pattern recognition and computer vision communities: boundary detection. Starting with simple, artificial images composed of two segments with discrete items characterized by just one or two features (in this case having target and distracters sharing one of them or both but differing by their combination), or more complex scene in which segmentation can produce different image interpretations such as in classical ambiguous pictures. A first preliminary, qualitative analysis, based on a limited set of about ten images and observers, is here discussed.