Abstract
Anatomical and perceptual findings suggest that color perception is a skill shared among the vast majority of organisms with advanced visual systems. Animals who possess it tend to exhibit better performance in a wide array of high and low level visual tasks such as segmentation and object identification. While most vertebra possess only dichromatic vision, humans and several other higher primates are trichromatic. The advantages of trichromatic vision have been studied extensively across a variety of tasks, and are quite apparent while comparing average trichromatic humans to their dichromatic “color-blind” counterparts. Such studies often focus on the performance gains trichromatic vision provides for tasks such as foraging, distinguishing between edible and non-edible plants, and even social interaction (where differentiation between different shades of skin pigmentation is advantageous). These studies may then proceed to claim that the spectral tuning of primate retinal pigments has evolved in order to optimally perform such tasks. Although studies indicate that small mutations in the genes encoding retinal pigments may produce large shifts in their peak response frequency - the retinal pigments of many higher primates share similar spectral sensitivities, despite environmental and behavioral differences between them. Therefore we examine the hypothesis that these pigments evolved to optimize a more general tasks, namely differentiating between spectra, rather than excel in a particular visual task. In this study standard evolutionary computation methods are employed over a large data-set of real-world spectral samples to discover optimal trichromatic filters for spectral differentiation. When simple, naturally plausible, constraints are applied (such as smoothness or an approximately Gaussian response curve) we find that human-like response profiles emerge naturally as optimal detectors. This strongly suggests that human and primate cone pigments are not only tuned to excel in specific visual tasks but rather are evolutionary optimized for differentiation between commonly occurring natural spectra.
Meeting abstract presented at VSS 2015