Skip to main content

Challenge 27: OCR - Extract Text from Images

Estimated Time

45 min | Cost: $1-3 (estimated) | Domain: Computer Vision Solutions (10-15%)

Exam skills covered​

  • Extract text from images using Azure AI Vision Read feature
  • Convert handwritten text to digital text
  • Process multi-page documents

Overview​

Azure AI Vision's Read feature (part of Image Analysis 4.0) extracts printed and handwritten text from images and documents. The text hierarchy:

Image → Blocks → Lines → Words (with bounding polygons and confidence)

Key characteristics:

  • Supports 164+ languages for print, 9 languages for handwriting
  • Handles rotated, skewed, and noisy text
  • Returns bounding polygons for each text element
  • Synchronous API for single images

For multi-page PDFs, use the Azure AI Document Intelligence Read model instead.

Prerequisites​

  • Azure subscription
  • Azure AI Services resource
  • Python 3.9+ or .NET 8
  • Package: azure-ai-vision-imageanalysis (v1.0+)

Implementation​

Task 1: Extract Printed Text from Images​

import os
from azure.ai.vision.imageanalysis import ImageAnalysisClient
from azure.ai.vision.imageanalysis.models import VisualFeatures
from azure.core.credentials import AzureKeyCredential

client = ImageAnalysisClient(
    endpoint=os.environ["AZURE_AI_ENDPOINT"],
    credential=AzureKeyCredential(os.environ["AZURE_AI_KEY"])
)

# Extract text from an image URL
image_url = "https://learn.microsoft.com/azure/ai-services/computer-vision/media/quickstarts/presentation.png"

result = client.analyze_from_url(
    image_url=image_url,
    visual_features=[VisualFeatures.READ]
)

if result.read:
    print("Extracted Text:")
    print("-" * 40)
    for block in result.read.blocks:
        for line in block.lines:
            print(f"  Line: '{line.text}'")
            print(f"    Bounding polygon: {line.bounding_polygon}")
            
            # Access individual words with confidence
            for word in line.words:
                print(f"      Word: '{word.text}' (confidence: {word.confidence:.4f})")

Task 2: Extract Handwritten Text​

# Read handwritten text from a local image
with open("handwritten-note.jpg", "rb") as f:
    image_data = f.read()

result = client.analyze(
    image_data=image_data,
    visual_features=[VisualFeatures.READ]
)

if result.read:
    print("Handwritten Text Extracted:")
    for block in result.read.blocks:
        for line in block.lines:
            # Check confidence - handwriting often has lower confidence
            avg_confidence = sum(w.confidence for w in line.words) / len(line.words)
            confidence_indicator = "✓" if avg_confidence > 0.8 else "?"
            print(f"  [{confidence_indicator}] '{line.text}' (avg conf: {avg_confidence:.3f})")

Task 3: Process Multi-Page Document with Document Intelligence​

from azure.ai.documentintelligence import DocumentIntelligenceClient
from azure.ai.documentintelligence.models import AnalyzeDocumentRequest
from azure.core.credentials import AzureKeyCredential

# For multi-page documents, use Document Intelligence Read model
doc_client = DocumentIntelligenceClient(
    endpoint=os.environ["AZURE_AI_ENDPOINT"],
    credential=AzureKeyCredential(os.environ["AZURE_AI_KEY"])
)

# Analyze a multi-page PDF
with open("multi-page-document.pdf", "rb") as f:
    poller = doc_client.begin_analyze_document(
        "prebuilt-read",
        body=f,
        content_type="application/pdf"
    )

result = poller.result()

print(f"Document contains {len(result.pages)} pages")
for page in result.pages:
    print(f"\n--- Page {page.page_number} ({page.width}x{page.height} {page.unit}) ---")
    for line in page.lines:
        print(f"  '{line.content}'")

# Access full content as continuous text
print(f"\nFull content:\n{result.content[:500]}")

Expected Output​

Extracted Text:
----------------------------------------
  Line: 'Azure AI Services'
    Bounding polygon: [{'x': 54, 'y': 28}, {'x': 403, 'y': 26}, ...]
      Word: 'Azure' (confidence: 0.9980)
      Word: 'AI' (confidence: 0.9950)
      Word: 'Services' (confidence: 0.9970)
  Line: 'Computer Vision'
      Word: 'Computer' (confidence: 0.9920)
      Word: 'Vision' (confidence: 0.9910)

Handwritten Text Extracted:
  [✓] 'Meeting notes - January 2024' (avg conf: 0.892)
  [?] 'discuss quarterly goals' (avg conf: 0.734)
  [✓] 'Action items below' (avg conf: 0.856)

Document contains 3 pages
--- Page 1 (8.5x11.0 inch) ---
  'Annual Report 2024'
  'Executive Summary'

Break & fix​

ScenarioSymptomRoot CauseFix
No text detectedEmpty resultsImage too small or low qualityMin 50x50 px; ensure adequate resolution (300 DPI for print)
Wrong language detectedGarbled textAuto-detection failed for rare scriptsSpecify language parameter in request
Low word confidenceUncertain resultsHandwriting quality or unusual fontsAccept lower thresholds for handwriting; preprocess image
413 Request Entity Too LargeFile rejectedImage exceeds 20MB limitCompress or resize image before submission
Bounding polygons incorrectMisaligned boxesImage rotation not detectedUse auto-rotation or preprocess to correct skew

Knowledge Check​

1. What is the text hierarchy returned by the Azure AI Vision Read feature?

2. What is the maximum image size supported by the Image Analysis Read feature?

3. For multi-page PDF documents, which Azure service should you use for text extraction?

4. How does the API indicate uncertainty in handwritten text recognition?

5. What Content-Type header should you use when sending a local image file for OCR?

Cleanup​

az group delete --name rg-ai102-vision --yes --no-wait

Learn More​