While edges mark the boundaries of objects or regions where image intensity changes sharply along one direction, corners represent points where intensity changes significantly in multiple directions. Think of the corner of a box, the tip of a star, or the intersection of two lines in an image. These points are often more distinctive than points along a straight edge.

What Makes a Point a Corner?

Imagine you have a small window or patch that you can slide across the image. Let's consider what happens to the appearance of the image content inside this window as you move it slightly in different directions:

Flat Region: If the window is over a large area of uniform color or texture (like a clear blue sky or a plain wall), sliding the window around won't change the content inside it very much. There are no significant intensity variations.
Edge: If the window is positioned over a straight edge (like the side of a building against the sky), sliding the window along the edge might not change the content much. However, sliding the window across the edge will result in a significant change as new intensities enter the window. The change is primarily strong in one direction (perpendicular to the edge).
Corner: Now, if the window is centered precisely on a corner, sliding it slightly in any direction (horizontally, vertically, or diagonally) will cause a noticeable change in the content within the window. This is because the intensity pattern changes along multiple directions radiating from the corner point.

This characteristic, the significant change in appearance within a local window regardless of the direction of movement, is the fundamental idea behind corner detection. Corners are points where the image gradient (the direction and magnitude of intensity change) varies significantly in more than one direction.

A conceptual diagram showing how a small analysis window behaves when moved over different image structures. Corners exhibit significant intensity changes within the window regardless of the direction of movement.

Why Are Corners Good Features?

Corners possess several properties that make them valuable features for computer vision applications:

Distinctiveness: Corners are often locally unique. Unlike a point on a smooth edge, a corner usually has a more specific intensity pattern around it.
Stability: They are generally stable under changes in illumination (brightness/contrast) and viewpoint (rotation, small perspective shifts). While the exact pixel values might change, the structure of the corner often remains recognizable.
Localization: Corners are well-localized points in the image, providing precise positional information.

Because of these properties, corners serve as excellent landmarks. They are frequently used in tasks like:

Image Stitching: Finding corresponding corners in overlapping images to create panoramas.
Object Tracking: Following the movement of corners on an object across video frames.
Structure from Motion (SfM): Reconstructing 3D scenes by tracking corner points across multiple viewpoints.
Object Recognition: Using constellations of corners as part of an object's description.

Understanding the concept of corners as points of high, multi-directional intensity change sets the stage for exploring specific algorithms designed to find them, such as the Harris corner detector discussed next.