Harris Corner Detector Algorithm

The Harris corner detector is an algorithm for finding corners, points where two edges meet and the image brightness changes sharply in more than one direction, within a digital image; corners of this kind make good stable landmarks for other computer vision tasks such as image stitching, tracking and three-dimensional reconstruction. At each pixel the algorithm builds a small matrix, often called the structure tensor, from the local derivatives of image intensity in the horizontal and vertical directions, then examines that matrix's eigenvalues; a flat region produces two small eigenvalues, an edge produces one large and one small eigenvalue, and a genuine corner produces two large eigenvalues, which the algorithm captures in a single corner response score without needing to compute the eigenvalues directly. Chris Harris and Mike Stephens published the method in 1988 as an improvement on Hans Moravec's earlier corner detector, and it remains a standard reference point-of-interest detector in computer vision.