One of the early stages in many computer vision pipleines is edge deection. This tries to identify pixels that sit on the boundary between parts of an image that have different characteristics, such as different objects, or different parts of the same object. Edge detection may use linear transformations such as gradient operators, Robert' operator, Sobel’s operator or the Laplacian operator followed by some sort of threshold. When deep neural networks are used in image recognition or processing, there may be no separate edge detction step, but early layers may train themselves to perform an equivalent function.
Defined on page 252
Used on Chap. 12: pages 244, 245, 246, 252, 253, 257, 258, 260, 263, 267, 276, 284, 285; Chap. 15: page 362; Chap. 22: page 540