People chose the measurements

Before deep networks took the lead, many image systems began with measurements designed by people.

A researcher might write procedures that respond to edges, corners, colour changes, or textures. These useful measurements are called features. A learning algorithm then used the chosen features to decide which object was present.

It was like giving a detective a prepared checklist: count the corners, measure the stripes, compare the colours. The detective could learn how to combine the answers, but people had already chosen which clues to collect.

The next system changed who shaped those measurements.