4.2.1. Azure AI Vision Service

Azure AI Vision (also called "Azure Vision in Foundry Tools") provides comprehensive image analysis capabilities. This is Azure's general-purpose computer vision service—think of it as your "one-stop shop" for most image analysis needs.

How Azure AI Vision works:

1. You create an Azure AI Vision resource in your Azure subscription
2. You send images to the service via REST API or SDK
3. The service returns structured analysis results (JSON)
4. No model training required—pre-trained models are ready to use

Capabilities include:

Image Description deep dive: The Image Description capability generates human-readable captions that describe the image content. Each caption includes:

• Description text: Natural language sentence
• Confidence score: How certain the model is (0.0 to 1.0)
• Multiple alternatives: Several possible descriptions ranked by confidence

Example output:

{
  "description": {
    "captions": [
      {"text": "a person walking on a sandy beach", "confidence": 0.92},
      {"text": "a beach scene at sunset", "confidence": 0.85}
    ]
  }
}

Specialized domain models:

Azure AI Vision supports two specialized domain models for categorizing images:

• Celebrities: Recognizes famous people (actors, politicians, athletes)
• Landmarks: Recognizes famous places (Eiffel Tower, Golden Gate Bridge)

⚠️ Exam Trap: Only celebrities and landmarks are specialized domain models. Azure Vision does NOT have specialized models for animals, cars, plants, or food. Questions asking about "specialized domain models" have celebrities and landmarks as correct answers.

What Azure AI Vision eliminates:

• Choosing a model (pre-trained models ready to use)
• Training a model (no training needed for pre-built capabilities)
• Evaluating a model (Microsoft handles model quality)
• Infrastructure management (Azure handles scaling)

What it does NOT eliminate:

• Azure resource provisioning (you must create a resource)
• Sending data for inference (you call the API)
• Processing the results (you handle the response)
• Paying for usage (charged per transaction)

When to use Azure AI Vision vs. custom models: