Edge Detection
Edge detection provides a more rigorous means than thresholding for initiating image segmentation. There is a large history of ad hoc edge detection algorithms, and this chapter aims to distinguish what is principled from what is ad hoc, and to provide theory and practical knowledge underpinning available techniques.
• the variety of template matching operators that have been used for edge detection—for example, the Prewitt, Kirsch, and Robinson operators.
• the differential gradient approach to edge detection—exemplified by the Roberts, Sobel, and Frei-Chen operators.
• theory explaining the performance of the template matching operators.
• methods for the optimal design of differential gradient operators, and the value of “circular” operators.
• tradeoffs between resolution, noise suppression capability, location accuracy, and orientation accuracy.
• outlines of more modern operators.
• the distinction between edge enhancement and edge detection.
In discussing the process of edge detection, this chapter shows that it is possible to estimate edge orientation with surprising accuracy within a small window—the secret being the considerable information residing in the gray-scale values. High orientation accuracy is of particular value when using the Hough transform to locate extended objects in digital images—as will be seen in several chapters in Part 2 of this volume.