Sharif Saleki¹, Kevin C. Hartstein¹, Peter U. Tse¹; ¹Dartmouth college

Object recognition is a central function of the brain’s ventral visual pathway. Many experiments have identified putative category-specific processing areas within human ventral temporal cortex. In particular, multivariate pattern analysis has revealed distributed representations of objects in Ventral Temporal (VT) cortex and Lateral Occipital Complex (LOC). Previous object recognition experiments have employed photographic images or photo-realistic line drawings as stimuli. However, humans also readily identify rough, quickly rendered, hand-drawn sketches of objects. Here, we selected stimuli from photographs and sketch drawings (collected from Google’s “Quick, Draw!” dataset) in animate (faces and body parts) and inanimate (natural and artificial objects) categories. Participants viewed static images of the photographs and dynamically drawn sketches in separate runs while brain responses were recorded via fMRI. Consistent with past research, multivariate pattern classification methods distinguished representations of different objects in photographic images in VT/LOC. Furthermore, above-chance performance of the classifier in decoding brain activity corresponding to rough sketches indicated the presence of information about their categories in object-processing areas. Cross-decoding between formats also yielded significant positive results, and the representational similarity followed previously reported structure of similarity in these regions. Together, we find shared representations of static photos and dynamically drawn sketches of objects, indicating a high-level representation of object category in the visual hierarchy that is largely feature-independent with respect to the presented stimulus format.