Amazon Rekognition Specialized2016年〜
An ML-based image recognition service that makes image and video analysis easy via API
What It Does
Amazon Rekognition is an ML-based service that analyzes images and videos through simple API calls. It offers object and scene detection, face detection and analysis, text extraction, inappropriate content detection, celebrity recognition, and more. No ML expertise is needed - just send an image to the API and get analysis results in JSON. You can also directly analyze images and videos stored in S3.
Use Cases
Auto-tagging and categorizing user-uploaded images, face matching for identity verification, automatic detection of inappropriate content on social media and community sites, automated visual inspection on manufacturing lines, and detecting and indexing specific scenes or objects in video.
Everyday Analogy
Think of an expert appraiser. Just show them a photo and they instantly identify objects, facial expressions, ages, text, and location features. They can analyze thousands of photos without fatigue, maintaining consistent accuracy.
What Is Rekognition?
Amazon Rekognition is AWS's image and video analysis service. Built on AWS's years of machine learning expertise, it provides pre-trained models accessible via API. Developers don't need to build or train ML models - just send an API request to get high-accuracy image analysis results. Two API sets are available: Image (for still images) and Video (for video content).
Key Features
Rekognition offers a rich set of analysis capabilities. Label detection identifies objects, scenes, and activities in images. Face detection and analysis estimates facial position, expression, age range, and gender. Face comparison determines whether two faces belong to the same person. Text detection reads text within images. Content moderation flags inappropriate images. With Custom Labels, you can even build your own object detection models using a small set of training data.
Video Analysis
Rekognition Video can analyze stored videos and live streaming footage. It supports person tracking, scene change detection, and pinpointing when specific objects appear. Integrating with Kinesis Video Streams enables real-time analysis of surveillance camera feeds for automated person detection and suspicious behavior alerts. Results include timestamps for easy navigation to specific scenes. For reference material when working on video analysis, books on Amazon can be helpful.
Pricing
Rekognition pricing is pay-per-use based on the number of images and videos analyzed. Image analysis is priced per 1,000 images with volume-based tiered discounts. Video analysis is billed per minute of processed video. The free tier includes 5,000 image analyses and 1,000 minutes of video analysis per month for the first 12 months. Custom Labels training and inference incur separate charges.
Things to Watch Out For
- Using facial recognition technology requires privacy and ethical considerations - check local regulations regarding intended use and notification requirements
- Analysis accuracy depends on image quality (resolution, brightness, angle), so define input quality requirements upfront and ensure quality through preprocessing
