Diffscriber: Describing Visual Design Changes to Support Mixed-Ability Collaborative Presentation Authoring

Visual Impairment Technologies (Screen Readers, Tactile Graphics, Braille)Universal & Inclusive DesignPhysicians, Nurses & CliniciansAssistive Technology SpecialistsFreelancers (Design, Writing, Translation)

Title of the Paper

Difscriber: Describing Visual Design Changes to Support Mixed-Ability Collaborative Presentation Authoring

Bibliographic Information

  • Subject Area: Human-Computer Interaction (HCI), Accessibility Technology
  • Keywords: Accessibility, Slide Authoring, Design Change Description, Visual Design, Multimedia Creation, Collaborative Tools, Mixed-Ability Teams, Blind or Visually Impaired Users

Research Background and Problem

  • Identified Issues or Challenges

    • Common slide authoring tools (e.g., PowerPoint, Google Slides) are not user-friendly for blind or visually impaired (BVI) users.
    • BVI users face difficulties in understanding and reviewing visual design changes in slides, especially when using screen readers, which only provide low-level visual layout information (e.g., (x, y) coordinates of elements).
    • BVI users often collaborate with sighted partners to create slides, but communication limitations frequently result in a lack of control over visual design.
  • Significance of the Research

    • Slide presentations are ubiquitous in educational and professional contexts, and improving the accessibility of slide authoring for BVI users is critical for ensuring equal participation.
    • Bridging the gap in visual capabilities between BVI and sighted users is essential for building inclusive technologies and workspaces.
  • Motivation and Related Work

    • Existing research primarily focuses on making slides “consumable” (e.g., adding alternative text to slides, improving the accessibility of visual content) rather than “editable.”
    • Mixed-ability collaboration emphasizes “interdependence,” and prior studies have explored accessibility tools in document editing contexts. However, the more complex demands of visual layout and style editing in slide design remain unaddressed.

Solution

  • Proposed Approach and System

    • The authors developed a system called Difscriber, which generates descriptions of slide design changes to help BVI users efficiently understand and review visual edits in slides.
    • Difscriber offers a multi-level navigation interface, including high-level change summaries, element-level change descriptions, and detailed visual style information for each element.
  • Innovative Contributions

    • Enables automatic detection and description of various types of slide changes, including layout, content, and style, supporting both asynchronous and synchronous collaboration.
    • Provides interactive, hierarchical descriptions, allowing users to flexibly navigate between high- and low-level descriptions and directly locate specific elements for editing.
  • Implementation Steps and Key Technologies

    1. Difference Detection: Uses embedding models (e.g., CLIP) to calculate similarity between slide versions, match elements, and classify change types (e.g., addition, deletion, style modification).
    2. Change Description Generation: Generates natural language descriptions based on templates, offering high-level summaries and fine-grained details.
    3. User Interface Design:
      • Slide Content List: Displays each element’s content and metadata (position, size, style, etc.).
      • Change Description List: Categorizes change information by layout, content, and style.
      • Edit Mode: Allows users to directly modify text content.
    4. Algorithm Optimization: Introduces determinant alignment (Itermax) and optical character recognition (OCR) to address matching errors and improve the accuracy of many-to-many text-image content matching.

Research Outcomes

  • Specific Results

    • In technical evaluations, Difscriber achieved an element matching accuracy of 77% and demonstrated good performance on presentation materials commonly used by BVI users.
    • User testing revealed that participants using Difscriber identified more changes (28.83 vs. 10.33) and provided more feedback (2.83 vs. 0.83) compared to those using only accessible slide content.
  • Advantages Over Existing Solutions

    • Compared to existing slide tools (e.g., PowerPoint), Difscriber enhances BVI users’ ability to perceive visual content and changes.
    • Facilitates efficient understanding of slide design changes and promotes collaboration and design dialogue between BVI and sighted users.
  • Experimental and Evaluation Results

    • Experiments involving six BVI users and two sets of slide designs provided systematic data support.
    • Users rated Difscriber significantly higher than accessible slides alone (6.33 vs. 3.33, p < 0.01).
  • Limitations and Future Directions

    • Limitations: The system currently does not support complex changes across multiple slides (e.g., merging and splitting) and relies heavily on automatically generated alt-text.
    • Future Work:
      • Cross-Platform Support: Extend functionality to other slide tools (e.g., PowerPoint).
      • Change Expansion: Enable summarization of changes across multiple slides.
      • Real-Time Enhancements: Develop real-time change stream descriptions for synchronous editing.
      • Enhanced Descriptions: Explore higher-quality evaluation methods tailored for design experts.

Quick Actions

Share

Share this page

ios_share

https://hci.top/en/papers/uist/84964/2022

AdRecommended

Learn AI Coding at CodeNow

open_in_newOpen DOI Link
DOI: https://doi.org/10.1145/3526113.3545637
At a Glance

Paper Snapshot

fact_check
dataset
Source
UIST
calendar_month
Year
2022
emoji_events
Award
No award tagged
group
Authors
4 authors
sell
Subtopics
Visual Impairment Technologies (Screen Readers, Tactile Graphics, Braille), Universal & Inclusive Design
work
Professions
Physicians, Nurses & Clinicians, Assistive Technology Specialists, Freelancers (Design, Writing, Translation)
article
Content Status
Full text indexed
hub
Related Papers
4 related papers