Understanding recognition from the use of visual information

We propose an approach that allows a rigorous understanding of the visual categorization and recognition process without asking direct questions about unobservable memory representations. Our approach builds on the selective use of information and a new method (Gosselin & Schyns, 2000, Bubbles) to depict and measure what this information is. We examine three face recognition tasks (identity, gender, expressive or not) and establish the information responsible for recognition performance. We compare the human use of information to ideal observers confronted to similar tasks. We finally derive a gradient of probability for the allocation of attention to the different regions of the face.