Amazon Rekognition

UpFreeOpen Sourcecloud by Amazon Web Services★62· JavaScript· MIT

Amazon Rekognition is image and video analysis — object detection, face recognition, text extraction, and content moderation.

Visit site ↗Source ↗Health checked 14h ago

Use it when

Pretrained models work out of the box — no training data needed

Watch for

Face recognition accuracy depends on lighting, angle, and age

First check

Call Rekognition.detectLabels / detectFaces / detectText for sync analysis. Video uses startLabelDetection as an async job that writes results to S3.

Auth

api_key

CORS

HTTPS

Yes

Signup

Required

Latency

47 ms

Protocol

REST

Pricing

paid

Stars

Uptime · 30-day window

Probes: 30Uptime: 100%Avg latency: 49ms

GitHub activity

★ 62JavaScriptMIT17 open issuesLast commit 155d ago

About this API

Rekognition is the entry point to AWS computer vision — pretrained models for general object detection, face detection and comparison, text extraction, content moderation, celebrity recognition, landmark recognition, and activity detection. Image analysis is synchronous (millisecond latency); video analysis runs as async jobs that write results to S3 with SNS notifications.

Face recognition has several usage patterns: detectFaces returns attributes (age, emotion, eyewear) without identifying who; compareFaces verifies whether two faces match; indexFaces + searchFaces build a face collection for 1:N search. Custom Labels (2019) lets you upload tens to hundreds of labeled images, AWS auto-trains a domain-specific model (e.g. detect manufacturing defects), and you call an inference API. Several jurisdictions regulate face recognition (Illinois BIPA, Washington State, post-IBM-exit legislation in many places); evaluate compliance before deploying.

What you can build

1Moderate user-uploaded content on UGC platforms
2Cluster and search faces in photo libraries
3Extract text from images (OCR)
4Detect celebrities, landmarks, and activities in video

Strengths & limitations

Strengths

Pretrained models work out of the box — no training data needed
Video API samples frames to control cost
Custom Labels lets you train domain-specific models with small datasets

Limitations

Face recognition accuracy depends on lighting, angle, and age
Some jurisdictions (e.g. parts of the US) regulate face recognition use

Getting started

Call Rekognition.detectLabels / detectFaces / detectText for sync analysis. Video uses startLabelDetection as an async job that writes results to S3.

FAQ

What is the Rekognition free tier?+

First 12 months: 5,000 images and 1,000 minutes of video per month.

How accurate is it?+

General object detection and text recognition approach commercial state-of-the-art. Face recognition is strong under standard conditions but degrades with lighting and angle. Use human review for high-stakes cases.

Does it support Chinese?+

detectText supports Chinese OCR. Other models output labels primarily in English.