ColorAssist: Perception-Based Recoloring for Color Vision Deficiency Compensation.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Image Processing

Background:

• Existing image enhancement methods often overlook individuals with Color Vision Deficiency (CVD).
• Current CVD compensation techniques lack rigorous validation by CVD individuals and struggle with balancing contrast and naturalness.
• This leads to suboptimal image quality for a significant global population affected by CVD.

Purpose of the Study:

• To develop a large-scale, CVD-individual-labeled dataset (FZU-CVDSet) for rigorous validation.
• To create an effective and perceptually accurate image recoloring algorithm (ColorAssist) for individuals with CVD.
• To address the limitations of existing methods in contrast enhancement and naturalness preservation for CVD.

Main Methods:

• Development of FZU-CVDSet, a novel dataset labeled by individuals with CVD.
• Introduction of ColorAssist, a CVD-friendly image recoloring algorithm.
• Design of perception-guided feature extraction and diffusion transformer modules for efficient recoloring.

Main Results:

• ColorAssist demonstrates superior performance in aligning with CVD visual perceptions compared to state-of-the-art methods.
• Comprehensive experiments on FZU-CVDSet and hospital-based subjective tests validate the algorithm's effectiveness.
• The proposed method achieves a better balance between contrast enhancement and naturalness preservation for CVD individuals.

Conclusions:

• ColorAssist represents a significant advancement in image enhancement for individuals with Color Vision Deficiency.
• The FZU-CVDSet dataset provides a crucial resource for future research and validation in CVD image processing.
• This work paves the way for more inclusive and perceptually accurate image enhancement technologies.