Independence from Context Information Provided by Spatial Signature Learning in a Natural Object Localization Task

Abstract

Although computational sciences have made considerable progress, artificial systems have not yet managed to duplicate the efficiency of human perceptive systems. This superiority may come from the human capacity to integrate world knowledge within perceptive processes. This study assesses the importance of context in a task where a target is to be located within a natural image. The use of context information is probably not the only reason for the high performance level of our perceptive system. Context information is modified using a jumbling process. Image quality is altered by blur or time constraint. The results show that, even if subjects are disturbed by image modifications, they are able to develop perceptive strategies used to selectively extract object distinctive features helping differentiate it from its surrounding.