Fig. 1: Analysis pipeline for a single voxel.

Top, full model. Bottom, constrained model. Natural scene images were converted to grayscale and passed as input to a steerable pyramid (shown here with only four orientations and three spatial frequencies for visualization purposes). Each filter yielded an energy response that was sampled by the voxel pRF, resulting in a scalar output for each filter. Linear regression was performed on the response amplitudes observed for each voxel with filter output values as predictors. This procedure yielded a set of model weights, which were subsequently used to simulate responses to a range of gratings in order to determine the voxel’s preferred orientation and spatial frequency. The constrained model (bottom) was identical to the full model except for an additional step of summing model outputs across orientations. As a result, the constrained model yielded a weight for each spatial frequency filter but enforced equal contribution across orientations. Example images shown here were created by the authors for illustration only and were not used in the study.