Evaluation of Automated Image Descriptions for Visually Impaired Students
Project Overview
The document explores the implementation of generative AI in education, focusing on automated image description systems designed to enhance accessibility for visually impaired students. It underscores the importance of providing accurate and comprehensive image descriptions to support effective learning. Through the use of structured templates, the study developed a method for generating these descriptions and utilized an evaluation questionnaire to assess their quality. The findings reveal that while the system successfully creates descriptions for simpler images, such as bar and pie charts, it struggles with more intricate diagrams, indicating a need for further refinement in these areas. Overall, the document highlights the potential of generative AI to improve educational accessibility, while also identifying challenges that must be addressed to fully realize its benefits in more complex visual contexts.
Key Applications
Automatic image description system
Context: Educational resources for visually impaired students
Implementation: Templates for image descriptions were created based on expert input and assessed using a structured evaluation questionnaire.
Outcomes: The generated descriptions showed potential for usefulness, especially for simpler diagrams. Some descriptions scored comparably to expert-generated descriptions.
Challenges: Descriptions for complex diagrams (e.g., node-link diagrams) scored lower, indicating a need for improvement.
Implementation Barriers
Technical/User Engagement
Current automated image descriptions do not consistently provide complete or structured results. There is a need for evaluation from visually impaired users to confirm the effectiveness of the descriptions.
Proposed Solutions: Utilizing structured templates derived from expert guidelines to improve the quality of descriptions. Future studies should include assessments involving visually impaired users.
Project Team
Anett Hoppe
Researcher
David Morris
Researcher
Ralph Ewerth
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Anett Hoppe, David Morris, Ralph Ewerth
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai