A research thesis on why data visualizations remain inaccessible to blind and visually impaired users, and what design and technical approaches can close that gap.
Most digital data is designed to be seen: charts, graphs, maps, and dashboards rely on spatial layout and color to communicate meaning. Screen readers and braille displays handle plain text reasonably well, but they largely fail once content becomes graphical — a scatter plot might be announced as "a chart with dots," which conveys almost nothing. WCAG, the primary web accessibility standard, gives strong guidance for text but offers little direction for making data visualizations themselves accessible, leaving this mostly unaddressed in practice.
- Reviewed data accessibility across three real-world contexts: home use, public spaces, and workplace settings
- Categorized data types (plain text, structured data, graphical data) since each requires a different accessibility strategy
- Surveyed existing assistive technologies: screen readers, braille displays, sonification, and tactile/3D-printed graphics
- Compared accessibility design methodologies: Universal Design, Cognitive Accessibility, Semantic Web approaches, and Multimodal Interfaces
- Conducted a technical deep dive into two specific systems: ARIA (the W3C accessibility standard) and SEE, a framework using RDF/RDFa/OWL to semantically annotate webpages for blind users
- Reviewed Drishti, a wearable device using a CNN (CaffeNet, trained on ImageNet) to describe a user's physical surroundings in real time
This is a literature review and design analysis thesis rather than a built-and-benchmarked system, so there are no accuracy/F1 metrics to report. The concrete outcomes are:
- Identified a specific, documented gap in WCAG: no clear guidance exists for accessible data visualizations
- Assessed SEE's semantic-annotation approach as more scalable than manual accessibility fixes, since it targets a page's underlying meaning rather than requiring per-visual manual description
- Evaluated ARIA's real-world adoption (BBC News, Spotify) alongside its practical limitations (steep learning curve, dependent on correct developer implementation)
- Established Drishti as a working proof of concept for real-time, CNN-based scene description outside the browser
- Proposed concrete future directions: AI-assisted ontology generation, deeper per-user personalization, and early exploration of AR/VR for data accessibility
Web standards: RDF, RDFa, OWL, ARIA, WCAG Deep learning: CNNs, CaffeNet, ImageNet Assistive technology reviewed: screen readers, braille displays, sonification, tactile graphics, 3D printing Languages/tools referenced in reviewed systems: Python, HTML, OpenCV, Tesseract OCR
This repository contains the thesis document, not executable code.
Author: Khushbu Patil — MSc Computer Science, Ca' Foscari University Venice Supervisor: Prof. Fabio Pittarello Contact: khushbu1207@gmail.com | linkedin.com/in/khushbupatil07 | github.com/PatilKhushbu