| Title |
Color Transformation AI Model for Color Vision Deficiency |
| Authors |
박성범(Seongbeom Park) ; 박노경(Nokyung Park) ; 백장현(Janghyun Baek) ; 정유진(Yujin Jeong) ; 정혜지(Haeji Jung) ; 김진규(Jinkyu Kim) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.11.80 |
| Keywords |
Color vision deficiency; Color transformation AI |
| Abstract |
Colors play a vital role in objects as they not only represent their unique characteristics but also facilitate differentiation and information conveyance. Color Vision Deficiency (CVD) is a condition resulting from the malfunction of retinal cone cells, leading to an inability to discern certain or all colors, for which no precise medical treatment exists. Utilizing an AI model for color transformation in individuals with CVD offers the advantage of easily adjusting the degree of color correction based on personal preferences, thanks to models trained on extensive datasets. However, the scarcity of suitable training data hinders the active pursuit of color transformation research using AI models in the context of CVD. Therefore, we propose two approaches to design and implement AI models for color transformation in individuals with CVD using limited data. The first approach involves utilizing CycleGAN’s unpaired image training to partition the dataset and train the model, addressing the lack of paired data in the CVD domain due to labeling complexities. The second approach explicitly defines and trains fine-grained color adjustment modules and transforms colors accordingly. To validate the effectiveness of the proposed methods, experiments were conducted using two datasets (Ishihara and Flowers-17), demonstrating the successful achievement of effective color transformation for individuals with CVD across all datasets. |