An AI-driven framework for perceived display spectra: The effects of dimming, observer age, and viewing distance

Abstract

Displays have become ubiquitous in modern society, serving as pervasive light sources that exert visual and non-visual effects on human physiology and behavior. Despite their widespread use and impact, a universal framework for characterizing perceived display light output across various viewing conditions still needs to be developed. This study introduces a novel, AI-driven framework for comprehensive perceived display light output characterization, accounting for the effects of observer age, viewing distance, and display dimming. The framework employs a deep neural network (DNN) trained on an extensive dataset of measured display spectra to predict spectral power distributions (SPDs) from RGB inputs. To simulate real-world scenarios, the DNN-predicted SPDs were transformed to account for viewing distance (36 cm–71 cm), display dimming (0–100 %), and observer age (1–100 years). The initial model achieved high accuracy (R2_avg = 0.99), maintaining robust performance even for challenging cases (R² > 0.94). Results show high accuracy in predicting photometric, colorimetric, and circadian measures. Future research will incorporate other parameters to the proposed framework.