Abstract
The ability of the human visual system to recognize and classify objects is one of the most impressive capacities and widely investigated topics in cognitive science. Such tasks are performed by the human brain with remarkable efficiency that exceeds that of any known artificial machinery. However, via what mechanisms are these individual objects classified and identified? Contemporary research relating to these problems provides a natural common ground for the convergence of two fields of study: functional brain imaging and computational modeling. Object-related areas in the ventral visual system in humans are known from imaging studies to be preferentially activated by object images compared with noise or texture patterns. It is unknown, however, which features of the object images are extracted and represented in these areas. Recent computational studies have shown the usefulness of selected object fragments as useful visual features for classification and recognition. In the present study we explored the extent to which the representation of visual classes used object fragments selected by maximizing the information delivered about the class. We tested fMRI BOLD activation of highly informative object features in low- and high-level visual areas, compared with non-informative object features matched for low-level image properties. Our results showed a significant correspondence between information content and activity in higher order object-related areas. That is, object-selective regions showed preferential activation for computationally acquired informative fragments as compared to non-informative ones - in the lateral occipital area (LO) and the posterior fusiform gyrus (pFs) activation by informative fragments was significantly higher for three object classes whereas activation in V1 was similar. Behavioral studies also revealed high correlation between performance and fragments information. The results show that an objective class-information measure can predict classification performance and activation in human object-related areas.
YL was funded by Charles Clore Fellowships. RM was funded by Benoziyo Center for Neurological Disorders, The Dominique Center and The Israeli Science Foundation Center of Excellence. SU was supported by Grant 7-0369 from the Israeli Science Foundation and EU Daisy Grant 015803.