What is computer vision?

Computer vision is the field of artificial intelligence that teaches machines to interpret and understand visual information from the world — photos, videos, camera feeds — in ways that mimic human sight but at machine scale. Instead of relying on explicit rules, modern computer vision systems learn from thousands or millions of labelled examples, a process that makes them a specialised application of machine learning. When you unlock your phone with your face, when a medical scan is checked for tumours, or when a self-driving car reads a stop sign, computer vision is doing the interpreting.

How computer vision learns to see

Traditional programming would try to tell a computer exactly what a cat looks like: whiskers, pointed ears, four legs. That approach breaks quickly because cats come in every colour, angle, and lighting. Computer vision flips the problem. You feed the system a large collection of cat pictures, each labelled “cat”, and it works out the patterns for itself. This training process uses neural networks — layers of mathematical operations that adjust based on the examples they see.

The same technique works for any visual task. Show a system thousands of X-ray images, some with a fracture marked and some without, and it learns the subtle visual cues that indicate a break. Show it hours of street footage with pedestrians highlighted, and it learns to detect people in new footage. The key ingredient is the training data: the more varied and accurately labelled the examples, the better the system generalises to new situations.

Everyday uses of computer vision

Face unlock is probably the most familiar example. The camera on your phone captures an image, the computer vision model maps the unique geometry of your face, and it compares that map to the stored version — all in under a second. Medical imaging is another high-impact application. Algorithms can flag suspicious regions in mammograms, retinal scans, or CT slices, helping radiologists prioritise the cases that need a closer look.

Manufacturing lines use computer vision to inspect products for defects at speeds no human could match. Retail systems track inventory from security footage. Self-driving cars combine multiple camera feeds to recognise lanes, traffic signs, pedestrians, and other vehicles. In each case the computer is doing what humans do — interpreting a visual scene — but doing it consistently at enormous volume.

Why computer vision is a machine learning problem

Classic programming can handle things like “if pixel > 128, mark as bright”. But recognising a face or a tumour is not a simple threshold. The relationship between raw pixels and the thing you want to identify is too complex to capture in hand-written rules. Machine learning solves this by allowing the computer to discover the relevant patterns from examples. That is why computer vision is sometimes described simply as machine learning applied to images — the “vision” comes from learning rather than from programmed logic.

Limitations to keep in mind

Computer vision systems are only as good as their training data. If the labelled examples are biased — for instance, mostly light-skinned faces, or mostly clean indoor images — the system will perform worse on people or scenes it was not trained on. They can also be fooled by subtle changes in lighting, angle, or by images deliberately designed to confuse them (adversarial examples). And they do not “understand” the way humans do; a model can correctly identify a stop sign without knowing what stopping means. For high-stakes decisions — medical diagnoses, autonomous driving — computer vision is a powerful assistant, but it still needs human oversight to catch its mistakes.