A feature is a compact measurement computed from a local group of pixels that captures something meaningful about the content, such as the presence, strength, or direction of a pattern. It is not a pixel value; it is a summary derived from many pixel values.
Why compact measurements help
Two properties make features more useful than raw pixels. Compression: a small patch of pixels becomes a few numbers, so comparisons are cheaper and less noisy. Abstraction: a measurement such as edge strength stays roughly stable when overall brightness shifts, so the representation tolerates changes that would wreck a direct pixel comparison.
The same door, two exposures
Photograph a door in bright sun and again in dim indoor light. The raw pixel values in the two images differ enormously — the dim version may have every value cut roughly in half. But the boundary between the door and the wall is still a sharp change in brightness in both images. A feature that measures 'how abruptly brightness changes here' reports a similar value in both cases, while a raw pixel comparison reports a large difference that has nothing to do with the door.
Features are not a fixed list. Different tasks call for different measurements, and the choice of feature is a design decision that shapes what the system can and cannot notice.