Checkpoint Analysis 1.1.1 Text Alternatives - Images
⚠️ Not currently in production. This checkpoint is not part of the live product right now: the image classification models it relied on were discontinued. Integration is planned for an upcoming release.
Ally
This criterion establishes that all non-text content presented to the user must have a text alternative that serves the same function or purpose. For images, this means identifying what role they play in the interface to determine whether they require a description, can be marked as decorative, or are interactive elements that need an accessible name.
Programmatic Strategy in the Detector
The analysis consists of two complementary phases:
1. Visual Identification and Classification
- We use Computer Vision to detect elements on the page that look like images (not only
<img>, but also:- Embedded SVGs,
- Canvas elements
- From this detection, elements are classified according to their likely functional role in the UI.
2. Interactive Adjustment via Prompt
The preliminary result is passed to an LLM module that enriches the contextual reasoning.
The evaluator or QA can help adjust prompts to improve precision for:
- Usage context
- Visual hierarchy
- Communicative role of the element
This allows the classification to adapt to different design patterns without losing consistency.
Functional Image Classification
| Image Type | Characteristics | Treatment in the Detector |
|---|---|---|
| Decorative | - Explicitly marked as decorative (alt="", aria-hidden, role="presentation")- Do not convey relevant information - Are not interactive - Located in repetitive areas such as headers, backgrounds, separators | Should not be reported as an error. The markup is validated for correctness. If not properly marked → Suggest marking as decorative. |
| Informative | - Convey relevant content or context - Generally found in content areas - Usually accompanied by descriptive text or titles | Verified to have meaningful alt text. If missing or insufficiently descriptive → Report issue 1.1.1 Informational Image Missing/Insufficient Alternative Text. |
| Active (Interactive) | - Part of a control or action (<a>, <button>, elements with onclick, tabindex, role="button", etc.)- Used for navigation, opening, expanding, saving, submitting, etc. | Must have an accessible name. aria-label, alt, aria-labelledby, or visible text content is validated. If missing → Report as 4.1.2 Name, Role, and Value (plus 1.1.1 if applicable). |
Expected Output
The detector:
- Identifies images using computer vision.
- Classifies them according to their visual-contextual function.
- Validates the presence and quality of the alternative text.
- Reports only cases where the markup is incorrect, insufficient, or absent.
This ensures:
- Reduction of false positives on decorative images.
- Prioritization of errors with real usability impact.
- Consistency with manual audit practices.