Non-Visual Information Transmission for the Blind and Visually Impaired
Tactile and multi-modal AI interfaces that make visual information (color, art, text) accessible.
This line of work, carried out at Sungkyunkwan University’s H-Lab, develops assistive technologies that translate visual information into non-visual channels for blind and visually impaired users.
Contributions include a reconfigurable tactile platform for non-visual information transmission and ColorWatch, a color-perceptual spatial tactile interface (Jabbar et al., 2021) (Lee et al., 2020). On the multi-modal AI side, the AI-Ekphrasis framework uses transformer-based foundation models to retrieve fine-grained text descriptions for images, improving mean retrieval ranking to 8.2 versus a 15.5 baseline (Jabbar et al., 2022). Related work on automatically generating art captions for visually impaired audiences received a Best Paper Award (Lee et al., 2021), and the tactile color-display device developed alongside this project was granted a Korean patent (Cho et al., 2020).
References
2022
- ElectronicsAI Ekphrasis: Multi-Modal Learning with Foundation Models for Fine-Grained Poetry RetrievalElectronics, 2022
2021
- ElectronicsColorWatch: Color Perceptual Spatial Tactile Interface for People with Visual ImpairmentsElectronics, 2021
- SPIELanguage of gleam: impressionism artwork automatic caption generation for people with visual impairmentsIn Thirteenth International Conference on Machine Vision, 2021
2020
- ICTCColorWatch: Color-Tactile Interpretation to Improve Color Perception and Accessibility for People with Visual ImpairmentsIn 2020 International Conference on Information and Communication Technology Convergence (ICTC), 2020
- PatentTactile device for displaying color information2020Korean Patent Application No. 10-2020-0148650