Literally, a point that locates a feature. A point is a location. A
feature is something interesting. In
computer vision feature points
are often used as the basis for many higher level
algorithms.
Much early research concentrated on the task of automatically
identifying edge points in images. These are locations where
there are sudden transitions in brightness or colour. Edge detection
algorithms can locate these points and give an indication of the
strength and direction of the edge in the image. Similar algorithms
exist to find corner points, where edges change direction.
Identifying feature points is often considered as a key first step
towards building an abstract computer representation of what is in a
scene. Once found, edges and corners can be linked to gain an idea of
the shapes within a scene, followed by attempts to recognise objects.
With real world scenes the algorithms can be unreliable or easily
fooled. Often the computer is helped by introducing easy to identify
features into the scene. For example, motion capture for animation often
has an actor dressed in black with white reflective spots on key parts
of their body. The computer can follow the spots much more reliably.
For specialist applications other types of features may be interesting.
In satellite image processing or industrial inspection, parallel edges
are often interesting (representing roads or cables).