Machine Learning
Amazon Rekognition
Amazon Rekognition analyzes images and video for labels, text, moderation, faces, and other visual signals. It can filter or enrich multimodal inputs before a foundation model receives them.
Key points
- Image APIs are synchronous while many video operations are asynchronous
- Moderation labels include hierarchical categories and confidence
- Face comparison is probabilistic and requires threshold and policy decisions
When to use it
- Screen uploaded images before multimodal generation
- Extract visual labels to improve media retrieval metadata
Exam tips
- Choose Rekognition for managed vision analysis and Textract for document structure and fields
- Calibrate confidence thresholds by risk and retain human review for consequential identity or moderation decisions