Amazon Rekognition
Image and video analysis: objects, faces, text, moderation.
❓ What is it?
A computer-vision service that detects objects and scenes, reads text in images, finds and compares faces, spots celebrities and unsafe content, and tracks people in stored or streaming video — plus Custom Labels for your own object classes.
💡 Why does it exist?
Vision models are expensive to build and maintain. Rekognition gives production-grade detection through an API so teams add sight to applications in an afternoon, paying per image or video-minute.
⏱️ When should you use it?
Use it for content moderation, identity verification workflows (face comparison), searchable media libraries, and PPE or product detection; use Custom Labels when you must recognise classes unique to your business.
🗺️ Where does it fit?
Behind upload and streaming paths: S3 events trigger Lambda to analyse new images; Kinesis Video Streams feeds live video analysis; results usually land in DynamoDB or OpenSearch for querying.
🔌 How do you integrate it?
Call DetectLabels, DetectModerationLabels, CompareFaces, or DetectText with an S3 object or byte payload; for Custom Labels, label a small dataset, train in the console, and start the project's inference endpoint.
🧩 Commonly integrated with
🎯 Exam angle (AIF-C01)
- Image/video scenario words — "moderation", "faces", "objects in photos" — map straight to Rekognition.
- Low-confidence face matches sent for human review is the canonical Rekognition + Augmented AI (A2I) pairing.