ColorAssist: Perception-Based Recoloring for Color Vision Deficiency Compensation

Color Vision Deficiency (CVD) significantly impairs individuals' ability to perceive specific colors, leading to disruptions in their daily routines. To advance research on visual signal recoloring and enhance the visual perception of those affected, numerous recoloring algorithms have been introduced. However, most existing CVD recoloring algorithms face two primary shortcomings: Firstly, the datasets employed have not been validated by individuals with CVD. Consequently, the compensation results have not undergone extensive verification by those with CVD. Secondly, their performance still requires enhancement. To address these issues, we have developed the first large-scale Perception-based CVD (PCVD) database, comprising 3,386 paired images evaluated by individuals with CVD. Based on this PCVD database, we propose a novel perception-based recoloring algorithm, named ColorAssist. Within this framework, we have designed a CVD-guided feature extraction module that transforms CVD-related signals into CVD-specific features. Additionally, we introduce a CVD-guided diffusion transformer module for achieving CVD-friendly signal recoloring and perception. Extensive comprehensive qualitative and quantitative experiments conducted with CVD participants demonstrate that the proposed ColorAssist closely aligns with their visual perception, providing content that is not only more color-natural but also exhibits harmonious contrast.