Amazon Rekognition

What it is

Amazon Rekognition is a pretrained computer vision service for analyzing images and videos. You call an API (Application programming interface)A defined way for one program to ask another for data or actions, such as a set of HTTPS requests and responses. Every AWS service is controlled through its API, which the console, the CLI, and the SDKs call for you.Glossary entry and get back what is in the picture, with confidence scores; no ML (Machine learning)Software that learns patterns from example data to make predictions or decisions, instead of following rules written by hand. A model is first trained on data, then used to make predictions on new data, which is called inference.Glossary entry expertise is needed. It analyzes images sent in the request or stored in Amazon S3, and videos stored in S3.

Choose Rekognition when a question asks to detect objects, scenes, text, faces, people, or unsafe content in images or video, or to verify a person's identity from a face, without building a model. To train a model for your own specific objects, use Rekognition Custom Labels, or Amazon SageMaker AI for fully custom ML.

Capabilities

Labels: detect objects, scenes, activities, and concepts, such as "car", "beach", or "running".
Text detection: printed and handwritten text in images, such as street signs or captions. DetectText returns at most 100 words per image (and per video frame), so for documents, such as forms, invoices, and pages of text, use Amazon Textract.
Content moderation: detect inappropriate, unwanted, or offensive content (explicit, suggestive, violent) in user-generated images and videos.
Face analysis:
- Detect faces and their attributes, such as facial landmarks, emotions, glasses, and age range.
- Compare two faces with CompareFaces (one-to-one): the face in a source image against the faces in a target image, for example a selfie against an ID photo. Nothing is stored.
- Search a face collection (one-to-many): store known faces with IndexFaces, then find matches for a new face with SearchFacesByImage (or SearchUsersByImage), for example to identify an employee or find people in a photo library.
Face Liveness: checks that a user is physically present in front of the camera during a short video selfie, to stop spoofing during identity verification.
Celebrity recognition.
Personal protective equipment (PPE) detection: for example, whether workers wear face covers, hand covers, and head covers.
Custom Labels: train a model with your own labeled images to find objects, logos, and scenes specific to your business, such as your products on shelves or defects on an assembly line.
Stored video analysis: start an asynchronous job on a video in S3, for example StartLabelDetection. Rekognition publishes the completion status to an SNS topic, often read through an SQS queue, and you then fetch the results.

Availability changes

Streaming Video and Bulk Image Analysis are no longer available to new customers, effective April 30, 2026. Accounts that used them in the previous 12 months keep access. Older exam questions pair Rekognition stream processors with Kinesis Video Streams for live face search. Do not choose stream processors for a new design: AWS suggests sending frames from the stream to the Rekognition Image API instead. All other Rekognition features are unaffected.

Security, resilience, performance, and cost

Security

IAM controls access to Rekognition APIs and the S3 objects it reads.
Face collections store face vectors, a mathematical representation of each face, not the images. Treat face data as sensitive biometric data, and follow privacy laws and consent requirements.
Interface VPC endpointA private connection from a VPC to an AWS service or another endpoint service without crossing the internet. Gateway endpoints, for Amazon S3 and DynamoDB, are route table targets; interface endpoints are network interfaces that use AWS PrivateLink.Glossary entry (AWS PrivateLink) keep API calls off the internet. CloudTrail records API calls.

Resilience and performance

Rekognition is a managed, Regional service that scales with your requests. Run stored video jobs asynchronously, and use SQS to process completion notifications reliably.

Cost

Pay per image analyzed, per minute of video analyzed, and for stored face metadata. Custom Labels charges for training time and for each hour the model is running, so stop a Custom Labels model when it isn't needed.

Monitoring

CloudWatch metrics for API calls, errors, and ThrottlingLimiting how many requests a client can make in a period. A throttled request is rejected, often with HTTP status 429 or a throttling error, and should be retried later with backoff.Glossary entry; CloudTrail for auditing.

Rekognition or another option?

If the requirement is	Choose
Detect objects, faces, or unsafe content in images and video	Amazon Rekognition
Find our own logos or products in images	Rekognition Custom Labels
Match a selfie against one ID photo	Rekognition `CompareFaces`
Identify a face among many known faces	Face collection with `SearchFacesByImage`
Prove a user is physically present during sign-up	Rekognition Face Liveness
Extract text, forms, and tables from scanned documents	Amazon Textract
A fully custom computer vision model	Amazon SageMaker AI
Store and stream live video from cameras	Amazon Kinesis Video Streams

Reverse mindmap: when the question points here

Is it Amazon Rekognition?

Yes, when the question says
- Detect objects or scenes in images leads to Rekognition labels
- Flag explicit or violent user uploads leads to Rekognition content moderation
- Verify a user against one photo leads to CompareFaces (one-to-one)
- Find a person in a set of known faces leads to Collection, SearchFacesByImage (one-to-many)
- Stop photo or video spoofing at sign-up leads to Face Liveness
- Check workers wear safety equipment leads to PPE detection
- Our own products or logos in images leads to Custom Labels
Know how video works
- Analyze a video in S3 leads to Start job, completion to SNS
- Live camera streams leads to Kinesis Video Streams, frames to the Image API
No, when the question says
- Forms, tables, invoices, dense text leads to Textract
- Train any custom model leads to SageMaker AI

Key takeaways

Rekognition is a pretrained image and video analysis service: labels, text, moderation, faces, celebrities, and PPE.
Face features include detection, comparison (CompareFaces, one-to-one), search in collections (SearchFacesByImage, one-to-many), and Face Liveness.
Custom Labels trains models for your own objects; stop the model when idle to save cost.
Stored video analysis is asynchronous, with completion published to SNS.
Streaming Video and Bulk Image Analysis closed to new customers on April 30, 2026.
DetectText returns at most 100 words per image; for documents, use Textract instead.