Abstract
Color processing in the cortex can be understood as a set of mathematical operations on signals originating in the cone photoreceptors. Linear approximations to these operations have been instructive in the retina and LGN but less so in V1. Under a linear model, any set of stimuli that evoke the same response should lie on a plane in color space. This prediction holds irrespective of static output nonlinearities and thus can apply to complex as well as simple cells. We measured isoresponse surfaces for 118 V1 neurons in cone contrast space and found that 40% conformed to this prediction. Data from the remaining 60% were better fit by quadratic surfaces. Some quadratic surfaces were cup-shaped, indicating sensitivity to narrow regions of color space. Others were ellipsoidal, indicating sensitivity to all color directions. The principal axes of quadratic surfaces tended to be aligned with L−M, L+M, and S directions, showing that these directions provide a useful basis for describing the color tuning of V1 neurons. Our results demonstrate that cone signals combine nonlinearly in V1 and represent a step towards a complete description of the operations that characterize color processing in the cortex.