Individual Color Vision Prediction from Cone Ratios: A Computational Approach

Abstract Human color vision varies significantly between individuals due to differences in the biological structure of the eye. Specifically, the ratio of long-wavelength-sensitive (L) to medium-wavelength-sensitive (M) cone photoreceptorsthe specialized cells that detect colorranges from 1.1:1 to 16.5:1 across the population (Hofer et al., 2005) 1. Despite two decades of research documenting these individual differences, no computational system has successfully predicted individual color perception from biological cone ratio measurements. Current computer vision approaches rely on population-average models that do not account for the individual variation observed in laboratory studies of human perceptual responses (psychophysical studies) (Fairchild, 2013; Hunt et al., 2005) 11, 12. We developed a computational visual system capable of predicting individual color vision responses from anatomical cone ratio data. The system implements an individual scaling methodology where cone responses are weighted proportionally to each person’s L:M ratio relative to a universal baseline reference. We validated this approach through systematic testing against three independent experimental datasets: Stockman & Sharpe cone fundamentalsstandardized measurements of photoreceptor sensitivityachieving 6-12 nm spectral accuracy within individual variation tolerance (Stockman et al., 2000) 16, Neitz laboratory unique green measurements across 14 subjects (92.9% prediction success rate with 8.2 nm mean error) (Neitz et al., 2001) 3, and Greene et al. detection threshold experiments using identical stimulus conditions (Greene et al., 2012) 8. The scaling formula predicts color discrimination variations, unique hue perceptions, and detection sensitivities across the biological range of human L:M ratios. This system enables personalized color vision research, clinical applications for individual assessment, and display optimization based on measured biological parameters.