Paria Mehrani presents “Multiplicative modulations enhance unique hue representation in V4” at Neuroscience 2019
Venue: Neuroscience 2019, Chicago, IL
Paper: Multiplicative modulations enhance unique hue representation in V4
Abstract:
Which region in the brain represents unique hues is unknown. We introduce a hierarchical model inspired by neural mechanisms in the visual system for local hue representation and with computational simulations suggest that cells in V4 have the capacity to encode unique hues. Our network of single-opponent color and hue-selective cells differs from that of [1, 2] as it models cells in each of LGN, V1, V2, and V4 areas and explicitly reveals how the contributions of each participating area can lead to a hue encoding. Our network receives cone activations as input and gradually increases nonlinearities in terms of cone responses as observed by [3]. Specifically, single-opponent LGN responses are obtained by linearly combining cone activation. Half-wave rectification keeps V1 tunings similar to those of LGN cells [4] while nonlinear in terms of cone inputs. De Valois et al. [1] suggested that additive/subtractive modulation of cone-opponent cells with S-opponent cell responses rotates the cone-opponent axes to red-green and blue-yellow directions. To achieve this rotation in V2, in addition to single-opponent cells, we propose multiplicative modulations of V1 L- and M-opponent cell activations with V1 S-opponent responses. Multiplicative modulations increase nonlinearities and mix color channels. Moreover, unlike additive/subtractive modulations with little impact on tuning bandwidths, multiplicative modulations reduce tuning bandwidths. Finally, V4 responses are obtained by linearly combining V2 activations with weights determined based on tuning peak distances of V2 cells to the desired V4 neuron tuning peak. Our results indicate that multiplicatively modulated V2 cells play an important role in the representation of hues along intermediate directions in the MacLeod-Boynton diagram [5]. Similarly, these cells have substantial input weights compared to single-opponent V2 cells to V4 neurons selective to unique green and blue hues. Moreover, we observed a gradual decrease in distance of tuning peaks to unique hue angles reported by [6] from our model LGN to V4. Our results show that responses of our network neurons resemble those of biological color cells.