Access All Areas: Making Accessibility More Accessible
summary
Claim: most existing PDFs are not accessible and manual remediation doesn’t scale. Demonstration extracts images from failing PDFs, calls Azure Computer Vision for captions and tags, inserts alt text with iText to create and tag PDFs, and runs a PDF/UA checker. Outcome: automated captions sometimes mislabel (fox → grass), checks can report success while human review is still required, success rates around 50% for complex visuals. Scale problem: about 1 trillion PDFs online, so automation needs human-in-the-loop and better handling of tables, charts, language and reading order.
AI-written summary · about
chapters (36)
- 0:00Intro
- 0:24Who am I
- 1:05What is accessibility
- 2:03W3C
- 4:08Accessibility
- 4:41How can we help
- 5:16The challenge
- 6:46Easy to solve
- 7:17Tools
- 10:13PDF
- 10:40Xkcd
- 11:05Vision and Translate
- 11:26Demo
- 11:45Code
- 12:18PDFs are complicated
- 12:37Expand stuff
- 13:24Search
- 13:41Alt Description
- 14:16It Works
- 14:42Vision
- 16:19Wallaby
- 16:42Captions
- 17:23Tags
- 18:33Dogs
- 18:46Not perfect
- 19:36Remove break point
- 20:31Accessibility check
- 21:29The future
- 24:20Context is everything
- 25:43Any questions
- 26:32Title
- 28:50PDFs
- 29:19Documents
- 30:41Challenges
- 33:44Reading Order
- 36:16Next Generation PDF
related talks
53mAccess All Areas: Making Accessibility More Accessible
28mFrom Documents to UI: Designing an Inclusive Digital Experience
29mUnderstanding Accessibility
50mQuality Control for Accessible PDFs
36mMaking sure your PDF/UA PDF is actually PDFUA
38mAutomated vs. Manual Accessibility Testing: Finding the Right Balance
25mSecurity and Accessibility can be friends!
17m15 Million Opportunities — Automated Accessibility at Scale