What is computer vision?
Computer vision is the field of AI that enables computers to interpret and understand visual information from images and video — identifying objects, reading text, detecting faces, and making sense of what a camera sees. Where natural language processing teaches computers to work with words, computer vision teaches them to work with pixels: to look at a photograph or a video frame and extract something meaningful from it, rather than seeing an undifferentiated grid of colour values. This article covers the common tasks the field groups together, how it works at a high level, where it shows up in the real world, the cloud services built around it, and — treated seriously — the bias and privacy concerns that come with it, before finishing with where it appears in AI certification study.
The plain idea
To a computer, an image is just a grid of numbers — pixel values with no inherent meaning attached. Computer vision is the set of techniques that let a computer go from that raw grid to a meaningful interpretation: this is a car, this word says “exit”, this is the same face as in another photo. It is, in effect, teaching machines to “see” in the sense of extracting understanding from visual data, not merely storing it.
Common tasks, explained plainly
Like NLP, computer vision is an umbrella over several distinct tasks:
- Image classification — assigning a label to an entire image, such as “cat” or “invoice”.
- Object detection — locating and labelling multiple objects within a single image.
- Image segmentation — outlining the exact pixels that belong to each object, rather than just a bounding box.
- Optical character recognition (OCR) — reading printed or handwritten text out of an image.
- Facial recognition — identifying or verifying a specific person from their face.
- Image generation — producing new images, a generative-AI capability covered in our explainer on what generative AI is.
How it works, at a high level
Modern computer vision is built on deep learning: models — often built on neural-network architectures — learn to recognise visual patterns by training on very large numbers of labelled images, gradually learning to associate certain pixel patterns with certain labels. Our explainer comparing AI, machine learning and deep learning covers where this fits in the broader hierarchy; this article deliberately stays at the “what it does” level rather than the underlying architecture.
Real-world uses
Computer vision underpins a wide range of applications already in use: medical imaging systems that help clinicians spot patterns in scans, self-driving and driver-assistance systems that need to interpret the road in real time, quality-inspection systems on manufacturing lines that catch defects automatically, document-processing systems that extract data from scanned forms and invoices, retail applications from inventory tracking to checkout, and security systems built around detection and recognition.
Cloud computer-vision services
Cloud platforms offer managed computer-vision capabilities so teams do not need to train vision models from scratch. On AWS, Amazon Rekognition provides image and video analysis, including object and text detection and facial analysis. Azure AI Vision and Google Cloud Vision offer comparable managed capabilities under their own AI product families. As with any managed service, exact features and pricing change over time — check each provider’s current documentation for specifics rather than treating any particular feature as fixed.
The honest and important ethical dimension
Computer vision carries real ethical weight, and this deserves to be treated seriously rather than as a footnote. Vision models can inherit bias from their training data — performing less accurately on some skin tones, age groups or lighting conditions if those were under-represented during training, which is a documented and serious accuracy concern for facial recognition in particular.
Beyond accuracy, facial recognition raises genuine privacy and surveillance concerns: the same technology that unlocks a phone can, deployed differently, track individuals across public spaces without their knowledge or consent. This is why facial recognition specifically, more than most AI applications, faces active regulation and public debate in many jurisdictions. Our explainer on responsible AI covers this territory in more depth — it is not a solved problem, and any real-world deployment of computer vision, particularly facial recognition, deserves careful, deliberate consideration of these risks rather than treating the technology as ethically neutral.
Computer vision in certification study
Computer vision sits within the AI and machine-learning fundamentals and use-case coverage on the AWS Certified AI Practitioner (AIF-C01), including recognising common vision tasks and the responsible-AI considerations around them, such as bias and privacy in facial recognition. The AWS Certified Cloud Practitioner also introduces AI/ML services, including vision-related ones, at a broad, introductory level.
Original practice questions, timed mock exams and revision notes. No card, nothing to pay.