Aws Rekognition

AWS Rekognition provides a broad computer vision API: detect objects and scenes in images, analyze faces (age range, emotions, landmarks), extract text from images (OCR), detect explicit/suggestive content, recognize celebrities, and analyze video for activity detection. For agents in document processing: extract text from document images with DetectText. For identity verification: compare faces with CompareFaces. For content moderation: detect NSFW content with DetectModerationLabels. For structured data extraction: detect labels with confidence scores. Tight S3 integration — pass an S3 URI instead of uploading bytes for large files. Pay-per-image pricing (no fixed infrastructure cost). Confidence is docs-derived.

AWS SDK (boto3 for Python, AWS SDK for JS/Java/.NET) wraps the Rekognition REST API. Key operations: rekognition.detect_labels(Image={S3Object|Bytes}, MaxLabels=10, MinConfidence=70) returns detected objects/scenes. rekognition.detect_text(Image={...}) returns detected text blocks (LINE and WORD). rekognition.compare_faces(SourceImage={...}, TargetImage={...}) compares two faces. rekognition.detect_moderation_labels(Image={...}) returns content moderation labels. rekognition.detect_faces(Image={...}, Attributes=['ALL']) returns facial analysis. Video: rekognition.start_label_detection(Video={S3Object}) starts async job; rekognition.get_label_detection(JobId) polls status.

AWS IAM auth: AWS access key + secret key (environment variables AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY) or IAM role (recommended for EC2/Lambda). IAM policy: rekognition:DetectLabels, rekognition:DetectText, etc. — granular per-operation. No API key; credentials via AWS STS. Boto3/AWS SDK handles request signing (SigV4). No OAuth2. Recommended: IAM role for compute services; avoid long-lived access keys.

Standard AWS error responses (JSON with Code, Message). ThrottlingException with exponential backoff retry. InvalidImageFormatException for unsupported image types. ImageTooLargeException for images exceeding limits (5MB inline; S3 for larger). API call limits: per-region, configurable. Service quotas manageable via AWS Service Quotas. Rekognition SLA: part of AWS SLA (99.9% for most regions). Video analysis: async jobs time out after 48 hours if not retrieved.

docs.aws.amazon.com/rekognition provides comprehensive documentation — operation reference, code examples in Python/Java/Node.js/.NET, developer guides for each feature, and pricing calculator. Getting started: AWS account, boto3 installed, IAM credentials configured, first DetectLabels call in under 10 minutes. Python boto3 examples are comprehensive. AWS Free Tier: 5000 images/month free for first 12 months. AWS Rekognition Custom Labels for training custom classifiers. Community via AWS Developer Forums and StackOverflow.