The bear before the forest, but the city before the cars: Revealing early object/background processing

Abstract

Natural scene recognition can be seen as a process of decomposition (first the gist is extracted and then objects identities) or a progressive build up (spatial relationships between objects give rise to the global scene). Since the contribution of the background on object identification is still controversial, knowing the processing times for object and background recognition as well as their interactions could be crucial. Previous reaction time studies by our group using manual responses suggest parallel object and context processing with early interactions (from context on object categorization and from salient objects on context categorization, Joubert et al., 2007, 2008). In order to assess the very first step of this early interference, we used a choice saccade task where participants had to choose between pairs of natural scenes according to their global context or to the object they contained. Manipulating the nature of object (animal, vehicle, no object) and background (natural, man-made) relationships, we showed that choosing on the basis of the background can be done very quickly and efficiently (median RT = 217 ms and 72% correct for natural, 213 ms and 74% correct for man-made). Roughly comparable performance was observed in the object discrimination task, although we found a strong search asymmetry with an advantage for animals (181 ms, 81%) over vehicles (207 ms, 63%). Interestingly, and in contrast to previous manual response studies, we found no evidence for object/background interactions (in terms of congruency) at such extremely short processing times, suggesting that early interactions during the first feed-forward sweep do not take place in the first steps of cortical processing but rather in higher level cortical areas.