What is computer vision?

In plain English

To a computer, a photo is just a grid of numbers, one for the colour of each pixel. Computer vision is the field of AI that turns those numbers into understanding: “this is a cat”, “that sign says STOP”, “this person is waving”. Modern computer vision is almost entirely deep learning: neural networks trained on millions of labelled images learn to spot edges, then shapes, then whole objects.

Why it matters

Computer vision is one of the most widely used kinds of AI, often without people noticing. It unlocks phones with your face, reads number plates and cheques, sorts parcels, helps radiologists spot problems on scans, checks products on factory lines and lets cars see the road. The field’s big breakthrough came in 2012, when a deep neural network called AlexNet won the ImageNet image-recognition contest by a wide margin and kicked off the deep learning boom. Today’s multimodal chatbots combine computer vision with language, so you can show them a photo and ask about it.

An example

You point your phone camera at a restaurant menu in Japanese. Computer vision finds the text in the image and reads each character; a translation model then shows the menu in English on your screen, in place of the original.

What to watch out for

Vision systems can be fooled by unusual angles, lighting or small changes a person wouldn’t notice. Their accuracy also depends on their training data: some face recognition systems have made more mistakes with darker skin tones, which matters when they are used by police or for ID checks. And cameras plus AI raise real privacy questions about who is being watched and identified.

See how a neural network breaks a photo into edges and shapes in AI vs machine learning vs deep learning.