Inside AI · Topic 6 of 10

Computer vision

Computer vision is the part of AI that gets information from pictures and videos, by finding patterns in the numbers that make up each image.

📌 Key points

Computer vision helps computers get useful information from images and videos.
A digital image is a grid of pixels. Each pixel is stored as numbers. In a colour (RGB) image each pixel has three values, red, green and blue, each from 0 to 255.
Main tasks: image classification (what is in the picture), object detection (what objects there are and where), segmentation (which pixels belong to each object), face recognition and OCR (reading text).
Convolutional neural networks (CNNs) slide small filters across the image. Early layers find simple features such as edges and corners; later layers combine them into shapes and objects.
Vision models learn from huge numbers of labelled images. The ImageNet dataset has millions of them.
Uses include helping doctors read medical scans, sorting recycling, checking crops, reading number plates and helping drivers.
Limits: models can be fooled by unusual lighting, angles or small changes, can be less accurate for groups missing from their training data, and face recognition raises privacy concerns.

Pixel: One tiny square of colour in a digital image.
Image classification: Deciding what the main thing in a picture is, e.g. "cat".
Object detection: Finding objects in a picture and where they are, often with boxes around them.
Feature: A useful pattern in data, such as an edge, a corner or a shape.
CNN: Convolutional neural network: a neural network that uses filters to find features in images.
Filter: A small grid of numbers that slides across an image to detect a feature, such as an edge.
OCR: Optical character recognition: turning pictures of text into text a computer can edit.

How can a CNN recognise a handwritten "7"?

The image is a 28 × 28 grid of grey pixels, so the computer sees 784 numbers.
Early layers use filters to find simple features: a straight line across the top and a slanted line going down.
Later layers combine these features into the shape of a whole digit.
The output layer gives a score for each digit from 0 to 9, e.g. "7": 95%, "1": 3%.
The network answers "7", the digit with the highest score.

💡 Remember

Remember: a computer never "sees" a picture like you do. It only sees a grid of numbers and finds patterns in them.

Test yourself8 questions from a bank of 19. A new mix every time.