What is Amazon Rekognition? A plain-English explainer

What computer vision is

Computer vision is the broader field this service belongs to: teaching software to extract meaningful information from images and video, the way a person glances at a photo and immediately understands what’s in it (see our explainer on what computer vision is). Historically this required training specialised machine-learning models on large labelled datasets — a serious undertaking on its own. A managed computer-vision service exists precisely to remove that barrier, offering the capability through an API instead.

What Rekognition actually does

Rekognition offers a handful of distinct analysis capabilities, all delivered the same way: send an image or video frame, get structured results back.

• Object and scene detection — identifying what’s present in an image, from everyday objects to broader settings like "outdoor" or "beach".
• Text detection — reading text that appears within an image, such as a street sign or a label on packaging.
• Content moderation — flagging content that may be inappropriate or unsafe, so it can be reviewed or filtered before it reaches an audience.
• Facial analysis — detecting faces in an image and analysing attributes such as apparent expression, without necessarily identifying who the person is.

Common uses

These capabilities map onto a fairly consistent set of real applications: moderating user-uploaded photos and video on a platform before they go live, making a media or photo library searchable by tagging what’s in each image automatically, extracting text from scanned documents or signage as part of a document workflow, and verifying identity as one step in an onboarding process. In each case, the service is doing the pattern-recognition work a human reviewer would otherwise have to do manually at a scale that doesn’t make sense to staff by hand.

The part that deserves honesty: facial recognition ethics

Facial analysis and facial recognition are the most capable — and most contested — part of computer vision, and it would be dishonest to describe this service without addressing that directly. Facial-recognition systems, across the industry and not specific to any one vendor, have documented accuracy differences across demographic groups, meaning error rates are not evenly distributed by skin tone, gender or age. Getting that wrong has real consequences when the system’s output feeds into decisions about identity or access.

Beyond accuracy, there’s a separate and equally serious concern: surveillance. The same technology that verifies a person’s identity for a legitimate onboarding flow can, if deployed without oversight, be used to track people’s movements and identities without their knowledge or consent. Responsible use of this kind of service means human review of consequential decisions, transparency with the people being analysed, and real caution about where and how facial recognition is deployed — not just because it’s good practice, but because getting it wrong causes tangible harm to real people. Our explainer on responsible AI covers this territory in more depth, and it’s worth reading before deploying facial analysis in anything that touches real decisions about people.

How it fits and what it costs

Rekognition is typically one component in a larger pipeline rather than the whole application — its structured output usually feeds into a moderation queue, a search index, or a decision workflow built elsewhere. Like most managed AI services, it’s billed on a pay-for-use basis tied to the volume of images or video processed, rather than a fixed subscription — check AWS’s current pricing documentation for the specifics rather than relying on any number repeated here, since usage-based pricing details change.

Cert-study tie-in

Rekognition is a standard example used to introduce computer vision on AWS’s AI-focused certifications, and exams test the conceptual understanding covered here — what image and video analysis capabilities exist and when you’d reach for them — plus, increasingly, the responsible-AI judgement around sensitive uses like facial recognition. Our /revision lessons cover the AWS Certified AI Practitioner syllabus at that depth.