Computer Vision

Computer Vision (CV) is the domain of AI that helps machines see. It takes in images or videos, finds patterns in the pixels and works out what they show, so that the machine can act on it.

CV models do four main jobs. For an image with one object: classification and classification + localisation. For images with many objects: object detection and image segmentation.

A digital image is a grid of tiny squares called pixels. Resolution is how many pixels the image has. Each pixel is stored as numbers: one number from 0 to 255 in a grayscale image, and three numbers (red, green, blue) in an RGB colour image.

A feature is a small, distinctive part of an image that helps a computer recognise it. Flat areas look the same everywhere, edges look the same all along a line, but corners are unique, so they make the best features.

Convolution slides a small grid of numbers, called a kernel, over an image. At each position it multiplies the pixels by the kernel numbers and adds them up. Different kernels blur, sharpen or find edges.

A CNN is a neural network built for images. Its convolution layer finds features, ReLU keeps only the useful positive values, the pooling layer shrinks the result, and the fully connected layer makes the final decision.

With no-code tools such as Teachable Machine and Lobe, anyone can train an image classifier: create classes, give examples, train, test and export. Try it with the shape sorter below.