Investigating the Role of Top Down Contextual Priors on Object Search in Real World Scenes
Speaker: Prof. Aude Oliva , Dept. of Brain & Cognitive Sciences, MIT
Relevant URL: http://www.csail.mit.edu/events/eventcalendar/calendar.php?show
Behavioral studies have shown that human observers make extensive use of contextual scene information during object search in natural images. In a series of experiments, we investigated the influence of top-down contextual priors in object search by monitoring eye movements as participants searched real-world scenes for objects (e.g. a pedestrian, a man or a woman, a painting, a cup). In our first study, we show that a computational model that relies on top-down categorical priors (the identification of the scene as a street, a park, etc.) can predict the location of human eye movements when the target object is small and camouflaged in the scene. In a second study, we investigate the relative contributions of bottom-up saliency and top-down contextual priors in predicting eye fixations during object search. Finally, in a third study, we evaluate how the strength of contextual priors (e.g. the probability of the association between a scene context and the object presence or location) influences the different stages of object search: the initial glance at the scene, the search process itself and the process-ing of the target object. Altogether, the results suggest a key role for experience-dependent top-down influences during visual search and impose constraints on computational models of vision. Work in collaboration with Barbara Hidalgo-Sotelo, Naomi Kenner, Antonio Torralba.