Automating Tactile Graphics Translation
Richard E. Ladner, Melody Y. Ivory, Rajesh Rao, Sheryl Burgstahler, Dan Comden, Sangyun Hahn, Matthew Renzelmann, Satria Krisnandi, Mahalakshmi Ramasamy, Beverly Slabosky, Andrew Martin, Amelia Lacenski, Stuart Olsen, Dmitri Groce · 2005 · Proceedings of the 7th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '05) · doi:10.1145/1090785.1090814
Summary
This paper addresses the critical shortage of tactile graphics available to blind students in mathematics, science, and engineering (MSE) courses. Postsecondary science textbooks may have images on half their pages — sometimes up to 500 images per book — yet these graphical images are rarely available in tactile form. The researchers conducted an in-depth study of current tactile graphics translation work practices, surveying 51 professionals through an online questionnaire and conducting sixteen three-hour observation sessions with tactile graphics specialists in the Pacific Northwest, California, and Kentucky. They found that the translation process is labor-intensive, slow, and highly variable in quality. Specialists — predominantly female (93%), with a median age of 48 — work across school systems and transcription agencies using a fragmented workflow involving separate tools for image drawing, Braille text creation, and embossing. Based on these findings, the team designed the Tactile Graphics Assistant (TGA), an automated pipeline that integrates image classification (identifying image type such as bar chart, line graph, or diagram), text segmentation and OCR, image simplification (reducing colors, removing backgrounds, replacing textures), and layout adjustment for Braille paper constraints (17 by 52 inches). The pipeline processes an original graphic through letter box identification, text removal and Braille translation, color-to-texture replacement, and final rendering for a Tiger Embosser.
Key findings
The study revealed that no existing software application integrates all the steps required for automated tactile graphics translation. Popular tools like Adobe Photoshop, CorelDRAW, and Microsoft Word each handle only parts of the workflow and lack Braille text support. Surprisingly, Microsoft Word was the most prominent tool used by specialists, despite not being designed for graphics translation. The image classification system achieved promising accuracy using Gaussian filters, oriented edge detection, and machine learning: precision ranged from 0.93 to 1.00 and recall from 0.86 to 1.00 across different chart types. Text segmentation successfully identified letter bounding boxes using a plane sweep algorithm from computational geometry, enabling automated OCR and Braille translation of text within images. The team estimated that automated steps would take less than a minute per image, while human review of textures and error correction would average 15 minutes — dramatically reducing the current timeline where several hundred images could take 1 to 5 hours per image (200 to 1,000 total hours per textbook). The multi-disciplinary team of eighteen included blind student consultants who validated the quality and understandability of the produced tactile graphics.
Relevance
This work addresses one of the most persistent barriers in STEM education for blind students: the lack of timely access to graphical content. When a graduate student co-author had tactile access to only 80 of 461 images in a computer architecture textbook, their academic performance suffered directly. The Tactile Graphics Assistant represents a significant step toward making textbook graphics available at scale, potentially reducing translation time by an order of magnitude. For accessibility practitioners, the study offers valuable insights into the tactile graphics production ecosystem — the specialists, their tools, workflows, and pain points. The research also demonstrates the importance of including blind users not just as test subjects but as active team members (student consultants and co-researchers) who shape the technology being developed. The automated pipeline approach — classify, segment, simplify, layout, render — provides a reusable framework that could be adapted as image processing and machine learning technologies continue to advance.
Tags: tactile graphics · braille · STEM accessibility · image processing · machine learning · visual impairment · assistive technology · automation · accessible publishing