Purpose: To determine the extent at which the cortical responses to images is driven by contours, texture or sparseness. Methods: Functional and anatomical images were acquired with a Siemens Sonata 1.5T MRI. The stimuli consisted of natural images, which were binarized to ensure identical contrast properties across all stimulus conditions. From this image set, two new image sets were created by setting half of the full binary images to mean luminance according to the energy of a bank of local oriented Gabor filters (peak s.f. ~40 cycles/image). Thus, three set of images were used: 1) full binary natural images and trinary natural images containing minimal (2) or maximal (3) local oriented structure, corresponding to contour and texture information in the images, respectively. The different stimulus conditions were alternated in a block design. During each block, subjects performed 2IFC contrast-discrimination tasks on two stimuli of the same condition. This task focused and maintained the subjects' attention. Both stereotaxic and a region of interest analysis on visual areas (V1 to V3A/V4v) were performed, the latter were identified in a separate scanning session. Results and Discussion: FMRI responses in early visual areas (V1) were largest for the full images, which may correspond to the amount of energy present. In higher visual areas (beyond V3/VP), however, the fMRI response was strongest to the images containing mainly contours, which may indicate a sparse code for contours.

Supported by CIHR grant MOP-53346 to RFH.