Computer vision turns pixels into useful information. Core tasks include classifying an image, detecting and locating objects, segmenting outlines, reading text and tracking motion across video frames. It powers phone cameras, self-checkout, medical scans, factory inspection and the image understanding in multimodal chatbots.
How it works
To a computer, an image is a grid of numbers: brightness values for red, green and blue at each pixel. Early computer vision relied on hand-designed rules to find edges and corners. The big shift came in 2012, when a deep convolutional neural network called AlexNet sharply cut error rates in the ImageNet image recognition challenge. Convolutional networks scan small filters across the image; early layers learn edges and textures, and deeper layers combine them into shapes and objects. More recently, vision transformers split an image into patches and process them much like tokens of text. Common tasks include classification ('this is a cat'), object detection (drawing boxes around each cat), segmentation (outlining exactly which pixels are the cat) and optical character recognition, which reads text. Models are trained on large labelled datasets. Multimodal models connect vision to language, so you can upload a photo and ask questions about it. Video adds time, letting systems track objects and actions across frames.
Why it matters
Computer vision automates looking. Creators rely on it for auto-reframing, background removal, face tracking and searching footage by what is in it. Businesses use it to scan receipts, inspect products, count stock and verify identity. Professionals in health, construction and farming use it to analyse scans, sites and crops. Understanding its strengths and blind spots helps you choose tools wisely and handle privacy, especially where faces are involved.
Where you will see it
- Removing or replacing a video background without a green screen
- Scanning receipts and invoices into accounting software
- Spotting defects on a production line from camera images
- Searching a photo library for a person, place or object
Limits and common misconceptions
- Performance drops with poor lighting, unusual angles or scenes unlike the training data.
- Face recognition raises serious privacy, consent and legal issues, and has shown uneven accuracy across groups.
- Models can be fooled by small, carefully designed changes to images that humans would not notice.
How it differs from related terms
- Computer vision vs Multimodal AI Computer vision focuses on understanding images and video, while multimodal AI combines vision with text, audio and other inputs in one system.
- Computer vision vs Text to image Text-to-image generates new pictures from words; computer vision analyses existing pictures.
Tools that use it
Frequently asked questions
How does computer vision work?
It converts images into grids of pixel values and runs them through neural networks trained on many labelled examples. Early layers detect simple patterns such as edges; later layers recognise shapes, objects and scenes. The network outputs labels, boxes, outlines or text, depending on the task it was trained for.
Is computer vision the same as image recognition?
Image recognition is one part of computer vision. The wider field also covers object detection, segmentation, reading text, estimating depth and pose, tracking motion in video and reconstructing 3D scenes. Recognition answers what is in an image; other tasks answer where, how many and how it moves.
Related terms
- Bias (in AI) Systematic patterns in an AI system's outputs that unfairly favour or disadvantage certain groups, ideas or outcomes.
- Chatbot Software that holds a conversation in natural language, by text or voice, to answer questions, complete tasks or provide information.
- Context window The maximum amount of text, measured in tokens, that a language model can take in and work with in a single request, including its reply.
- Copilot An AI assistant embedded inside another app that uses what you are working on as context to suggest, draft or complete tasks.
- Deep learning Machine learning that trains neural networks with many stacked layers to learn patterns directly from raw data such as text, pixels or sound.