👁️
Computer Vision
AWS services for image and video analysis
⏱️ Estimated reading time: 18 minutes
Amazon Rekognition
Deep learning-based image and video analysis service that requires no ML experience.
- Identifies thousands of objects (vehicles, pets, furniture)
- Recognizes activities (running, dancing)
- Detects scenes (beach, city, sunset)
- Face detection: Location and attributes
- Face comparison: Similarity between two faces
- Face search: Search for people in collections
- Facial analysis: Estimated age, gender, emotions
- Identifies celebrities in images and videos
- Provides additional information
- Detects inappropriate or offensive content
- Classifies by categories (violence, drugs, etc.)
- Adjustable confidence levels
- Extracts text from images
- Supports printed and handwritten text
- Multiple languages
- Personal Protective Equipment
- Helmets, gloves, masks
- Useful for safety compliance
- Identity verification
- Security and surveillance
- Social media content moderation
- Media analysis
- Access control
- Visual search
Key Features
Object and Scene Detection
- Identifies thousands of objects (vehicles, pets, furniture)
- Recognizes activities (running, dancing)
- Detects scenes (beach, city, sunset)
Facial Recognition
- Face detection: Location and attributes
- Face comparison: Similarity between two faces
- Face search: Search for people in collections
- Facial analysis: Estimated age, gender, emotions
Celebrity Recognition
- Identifies celebrities in images and videos
- Provides additional information
Content Moderation
- Detects inappropriate or offensive content
- Classifies by categories (violence, drugs, etc.)
- Adjustable confidence levels
Text Detection (OCR)
- Extracts text from images
- Supports printed and handwritten text
- Multiple languages
PPE Detection
- Personal Protective Equipment
- Helmets, gloves, masks
- Useful for safety compliance
Use Cases
- Identity verification
- Security and surveillance
- Social media content moderation
- Media analysis
- Access control
- Visual search
🎯 Key Points
- ✓ Rekognition provides production-ready features without custom models
- ✓ Validate accuracy across demographic groups to avoid facial recognition bias
- ✓ Optimize images (resolution, format) to improve OCR and detection
- ✓ Consider video processing vs image cost tradeoffs
- ✓ Combine services (Rekognition + Textract) for heterogeneous pipelines
Amazon Textract
Service that automatically extracts text, handwriting, and data from scanned documents.
- Advanced OCR
- Printed and handwritten text
- Preserves document layout
- Identifies key-value pairs
- Extracts data from form fields
- Checkboxes
- Extracts tabular data
- Maintains row and column structure
- Processes complex tables
- Passports
- Driver's licenses
- Extracts structured information
- Identifies signatures in documents
- Useful for validation
- Searches for specific information in documents
- Natural language questions
- Example: 'What is the total amount?'
- Understands document structure
- No manual configuration required
- Handles multiple formats
- High accuracy
- Invoice processing
- Mortgage loan automation
- Medical records extraction
- Legal document digitization
- Insurance claims processing
- Customer onboarding (KYC)
Capabilities
Text Extraction
- Advanced OCR
- Printed and handwritten text
- Preserves document layout
Forms Analysis
- Identifies key-value pairs
- Extracts data from form fields
- Checkboxes
Tables Analysis
- Extracts tabular data
- Maintains row and column structure
- Processes complex tables
Identity Document Analysis
- Passports
- Driver's licenses
- Extracts structured information
Signature Detection
- Identifies signatures in documents
- Useful for validation
Queries
- Searches for specific information in documents
- Natural language questions
- Example: 'What is the total amount?'
Advantages over Traditional OCR
- Understands document structure
- No manual configuration required
- Handles multiple formats
- High accuracy
Use Cases
- Invoice processing
- Mortgage loan automation
- Medical records extraction
- Legal document digitization
- Insurance claims processing
- Customer onboarding (KYC)
🎯 Key Points
- ✓ Textract is ideal for extracting document structure; always validate complex tables and forms
- ✓ Preprocessing images (deskew, contrast enhancement) improves OCR
- ✓ Test with real documents and variants (languages, formats)
- ✓ Use Queries to extract specific fields at scale
- ✓ Manage PII and apply redaction when necessary
Other Vision Services
Amazon Lookout for Vision
Purpose: Detect defects in manufactured products
Features:
- Automated visual inspection
- Identifies anomalies and defects
- Training with few images (30+)
- Integration with production lines
Use cases:
- Manufacturing quality control
- Damaged product detection
- Assembly verification
Amazon Monitron
Purpose: Predictive monitoring of industrial equipment
Features:
- End-to-end predictive maintenance system
- Sensors for vibration and temperature
- Detects abnormal behavior in machinery
- Automatic ML without development needed
AWS Panorama
Purpose: Computer vision analysis at the edge
Features:
- Physical device (Panorama Appliance)
- Processes video from IP cameras locally
- Low latency
- Privacy (data doesn't leave the site)
Use cases:
- Real-time PPE verification
- People counting
- Manufacturing quality control
- Retail traffic analysis
Amazon Rekognition Video
Video Analysis:
- Activity detection
- Person tracking
- Real-time facial recognition
- Inappropriate content detection
- Sports event analysis
🎯 Key Points
- ✓ Lookout for Vision is useful for industrial inspection with few training images
- ✓ Monitron and Panorama cover edge and predictive maintenance scenarios
- ✓ Assess use-case sensitivity and latency to choose edge vs cloud
- ✓ Integrate inference pipelines with alerts and action systems
- ✓ Measure false positives/negatives in production and tune thresholds