Abstract
Human color perception is categorical. According to previous psychophysical studies, category has profound effects on color discrimination and memory. For example, two memory colors in different categories are more accurately discriminated than colors within an identical category (Uchikawa et al., 1996). Focal colors are remembered more accurately than non-focal colors (Heider, 1972). The mean difference between test color and recalled color increases with the delay time (Perez-Carpinell et al., 1998). Meanwhile, the activities of IT color-coding neurons differed depending upon whether the task demand is discrimination or categorization (Koida et al., 2007). Despite the accumulating evidences that suggest interaction among color category, memory and discrimination, there is no inclusive modeling study that accounts for how these different aspects of color processing emerge from the nervous system. In this study, we propose a computational model that explains the categorical effects in color perception and memory, assuming a two-dimensional space of hue and saturation. The current model considers two populations of color-coding neurons in visual cortex: ‘hue-selective neurons’ and ‘category-selective neurons’. In the former, preferred hues are homogeneously distributed, while in the latter the preferred-hues are concentrated around some points (focal colors). We modeled the recurrent processing between the two neural populations, demonstrating the iterative Bayesian estimation of the presented color information based on the previous observations. Our model reproduced the previously reported physiological and psychophysical phenomena, which include the properties of color-coding neurons in IT cortex during categorization/discrimination, and the categorical effects and the changes of saturation in color memory. These results suggest that perceptual biases found in color processing and task-dependent modulations of neural responses are explained as natural consequences of statistically optimal estimation.
This research is supported in part by the Japan Society for the Promotion of Science (JSPS) through its.