Skip to content

Analyze Images

This section of the Microsoft AI-102: Designing and Implementing a Microsoft Azure AI Solution exam covers analyzing images with Azure AI Vision. Below are study notes for each sub-topic, with links to Microsoft documentation, exam tips, and key facts

Select Visual Features to Meet Image Processing Requirements

📖 Docs: Azure AI Vision features

Overview

  • Azure AI Vision can extract a variety of features from images
  • Features include:
    • Tags
    • Objects
    • Categories
    • Descriptions
    • OCR (printed and handwritten text)
    • Spatial analysis

Key Points

  • Select features based on requirements
  • Multiple features can be combined in a single request
  • Feature choice affects cost and performance

Exam Tip

Watch for scenario questions mapping requirements → correct feature

Detect Objects in Images and Generate Image Tags

📖 Docs: Object detection

Overview

  • Object detection identifies entities within an image
  • Image tagging generates a set of descriptive labels

Key Points

  • Tags include confidence scores
  • Object detection provides bounding boxes
  • Can identify thousands of common objects

Use Case

Retail solution detecting products on shelves with bounding boxes

Include Image Analysis Features in an Image Processing Request

📖 Docs: What is Image Analysis?

Overview

  • Image processing requests specify which features to analyze
  • Request payload includes:
    • Image source (URL or binary data)
    • List of features

Key Points

  • REST API and SDKs available (Python, C#, Java, JavaScript)
  • Features requested determine output fields
  • Can batch multiple images in one request

Exam Tip

Remember that URL or binary data can be used as inputs

Interpret Image Processing Responses

📖 Docs: Image descriptions

Overview

  • Responses contain structured JSON results
  • Includes:
    • Tags with confidence
    • Object bounding boxes
    • Category hierarchy
    • Text regions for OCR

Key Points

  • Always check confidence thresholds
  • Low-confidence results may need filtering
  • Can integrate results with downstream apps (search, indexing, etc.)

Best Practices

Filter out results below a set confidence threshold for production apps

Extract Text from Images Using Azure AI Vision

📖 Docs: OCR - Optical Character Recognition

Overview

  • OCR extracts printed text from images and documents
  • Works with multiple languages
  • Provides text lines and bounding box coordinates

Key Points

  • OCR is asynchronous for large documents
  • Text can be returned in plain text or structured format
  • Often combined with Document Intelligence for advanced extraction

Use Case

Digitizing scanned contracts into searchable text

Convert Handwritten Text Using Azure AI Vision

📖 Docs: Vision Portal demo

Overview

  • Recognizes handwritten text in images and documents
  • Supports cursive and block-style handwriting
  • Returns extracted text and bounding boxes

Key Points

  • Accuracy depends on handwriting quality
  • Works best with clear, high-resolution scans
  • Can be used in note-taking or form digitization apps

Exam Tip

Keywords like handwriting or forms with writing → Azure AI Vision handwriting OCR

Quick‑fire revision sheet

  • 📌 Visual features: tags, objects, categories, descriptions, OCR
  • 📌 Object detection → bounding boxes, tagging → descriptive labels
  • 📌 Requests can use URL or binary input, features specified per request
  • 📌 Responses contain confidence scores, bounding boxes, structured JSON
  • 📌 OCR extracts printed text, Handwriting OCR extracts cursive/block text