Systematic Survey of AI Image Description for STEM Accessibility
A comprehensive review available on arXiv investigates 20 peer-reviewed studies concerning AI methods for interpreting STEM visuals, emphasizing accessibility for those who are blind or have visual impairments. Adhering to PRISMA guidelines and utilizing a ROBIS-based bias risk evaluation, the study examines the specific STEM visuals addressed, the AI and machine learning frameworks used, the datasets and metrics for evaluation, and the methods of interaction for providing descriptions. Findings indicate a transition from static, one-time solutions to more engaging and interactive systems. However, the overall research landscape is still disjointed, showing minimal real-world effects on users.
Key facts
- Survey examines 20 peer-reviewed studies
- Focus on AI-based description techniques for STEM images
- Targets accessibility for blind or visually impaired individuals
- Uses PRISMA methodology and ROBIS-based risk-of-bias assessment
- Analyzes STEM visual types, AI architectures, datasets, evaluation metrics, and interaction modalities
- Identifies shift from static to dynamic description approaches
- Research landscape is fragmented
- Limited real-world impact on users
Entities
Institutions
- arXiv