How to Use Azure Vision in Foundry Tools for Image Analysis and OCR?

Azure Vision in Foundry Tools provides ready-made computer vision capabilities that can be used directly from applications without building or training custom machine learning models.

1. Core features in Azure Vision (Foundry Tools)

Azure Vision includes several major capabilities:

• Image Analysis
  • Extracts visual features from images such as:
    • Objects
    • Faces
    • Adult/racy content
    • Auto-generated tags
    • Auto-generated natural-language descriptions
  • The Image Analysis 4.0 API is the latest version and adds:
    • Synchronous OCR (Read OCR) for images
    • People detection
  • Can be accessed via SDKs or the REST Analyze Image API.
• OCR (Optical Character Recognition)
  • Extracts printed and handwritten text from images, including:
    • Photos of street signs, products, posters, screenshots
    • Text-heavy images and multi-page PDFs (via the Document Intelligence Read OCR model for documents)
  • Azure Vision OCR in Image Analysis 4.0:
    • Fast, synchronous API optimized for non-document, “in-the-wild” images
    • Supports printed and handwritten text in many languages using script-based models
    • Returns text plus structure via the readResult section in the Analyze Image response.
• Face
  • Detects, recognizes, and analyzes human faces in images.
  • Supports scenarios such as identification, touchless access control, and face blurring.
• Thumbnail generation
  • Analyzes an image to find the area of interest and generates a high-quality thumbnail, optionally with a different aspect ratio than the original.

2. Integrating Azure Vision without custom ML models

Azure Vision is exposed as managed APIs and SDKs, so integration does not require ML expertise:

• Use client SDKs or REST APIs
  • Image Analysis 4.0 can be called via:
    • Client library SDKs (for example, .NET, Python, etc.)
    • Direct REST calls using the Analyze Image API (https://aka.ms/vision-4-0-ref).
  • For OCR with Image Analysis 4.0:
    • Call the Analyze Image API and include Read in the features query parameter.
    • Parse the readResult section of the JSON response for extracted text.
• Use SynapseML / Synapse Analytics (no ML model building)
  • In Azure Synapse Analytics with SynapseML, Vision can be used as Spark transformers:
    • AnalyzeImage – extract tags, objects, faces, adult content, descriptions.
    • OCR / RecognizeText – read text from images.
    • DescribeImage – generate human-readable descriptions.
    • ReadImage – used in the tutorial to run OCR and return results in a Spark DataFrame.
  • Example (OCR in SynapseML):

        df = spark.createDataFrame([
            ("<replace with your file path>/ocr.jpg", )
        ], ["url", ])
        
        ri = (ReadImage()
            .setLinkedService(ai_service_name)
            .setImageUrlCol("url")
            .setOutputCol("ocr"))
        
        display(ri.transform(df))
    

  • This pattern lets applications use Vision from data pipelines without training or deploying models.

3. Typical use cases

• Text extraction (OCR)
  • Extract printed and handwritten text from:
    • Product labels, street signs, posters, screenshots
    • Business documents, invoices, receipts, reports, letters, whiteboards (for text-heavy documents, use Document Intelligence Read OCR model).
  • Power scenarios like:
    • Knowledge mining and search
    • Business process automation (invoice processing, form intake)
    • Accessibility (screen-reading of images)
• Image analysis and tagging
  • Automatically tag images with objects and concepts.
  • Detect faces and adult/racy content.
  • Generate natural-language descriptions of images.
  • Example from SynapseML tutorial: analyzing a dog image returns tags like "dog", "outdoor", "fence", "bench".
• Thumbnail generation and content-aware cropping
  • Generate thumbnails that focus on the most important region (e.g., a person’s face) while fitting a required aspect ratio.
• Face-based scenarios
  • Detect, recognize, and analyze human faces for:
    • Identification and verification
    • Touchless access control
    • Privacy-preserving face blurring.
• Domain-specific content detection
  • Detect and identify domain-specific content in images (as demonstrated in the SynapseML tutorial).

Note: Real-time video analysis is not a primary Azure Vision scenario in this context. For advanced video analysis (transcription, translation, summarization), Video Indexer is recommended instead.

4. Limitations and prerequisites

From the provided context, the key considerations are:

• Service scope and when not to use Azure Vision
  • Use Azure Vision for:
    • Extracting printed and handwritten text from images and documents via OCR.
    • Analyzing images for objects, faces, and descriptions.
    • Detecting and analyzing human faces.
  • Do not use Azure Vision for:
    • Advanced video analysis (use Video Indexer).
    • Content safety moderation (use Content Safety in Foundry Control Plane).
    • Tasks already covered by large multimodal foundation models like GPT‑4o.
• API behavior and performance
  • Image Analysis 4.0 OCR:
    • Synchronous, optimized for non-document images and near real-time experiences.
  • Document Intelligence Read OCR:
    • Asynchronous, optimized for text-heavy PDFs and scanned documents.
    • Requires polling via a “Get Read Results” operation to retrieve results.
• Deprecation notice
  • Image Analysis 4.0 in Azure Vision in Foundry Tools is marked as deprecated and will be retired on September 25, 2028. After that date, calls to this service will fail.
  • Migration to alternatives outlined in the migration guide is recommended.
• Language support
  • OCR supports several languages for printed text.
  • Handwritten OCR is currently supported exclusively for English (for the legacy OCR described in the transparency note).
• General prerequisites
  • An Azure AI Foundry / Azure Vision resource and keys.
  • For SynapseML usage:
    • Azure Synapse Analytics workspace.
    • SynapseML library configured.
    • Linked service to the Azure AI service (ai_service_name in examples).

File formats, size limits, and region-specific availability are not detailed in the provided context; those details are typically documented per API in the service reference and should be checked there when designing production solutions.

References:

• What is Azure Vision in Foundry Tools?
• What is Image Analysis?
• Tutorial: Azure Vision in Foundry Tools
• OCR for images (version 4.0)
• Transparency note and use cases for optical character recognition
• Foundry Tools
• Choose an Azure AI image and video processing and generation technology