Canny Edge Detection

An edge is a place where image intensity changes rapidly. It can mark the side of an object, a cast shadow, text, texture or noise. John Canny’s detector was designed to find useful, well-localized edges while limiting extra responses. The familiar practical pipeline first smooths an image to reduce small noisy fluctuations, estimates the brightness gradient, thins candidate responses and links convincing segments with hysteresis thresholds. OpenCV documents these stages in its Canny tutorial. Smoothing trades noise suppression against fine-detail loss. A strong blur can remove a hairline crack along with sensor speckle. The gradient estimates both how large a local change is and its direction. Non-maximum suppression keeps points near local gradient peaks, creating thinner lines rather than broad bright bands. Two thresholds then distinguish strong edges from weaker candidates. A weak segment connected to a strong edge may survive, while an isolated weak response is rejected. The chosen thresholds and image scale change what appears. The output is not semantic understanding. A stripe painted on a wall can produce edges despite no physical boundary. A low-contrast product edge can be absent. Repeated texture may create many small curves, and shadows may shift with lighting. Preprocessing, camera exposure and color-to-grayscale choices affect results. Inspect edge maps on representative images rather than selecting parameters from one ideal photo. For a task such as page cropping or line measurement, add geometric checks: contour shape, expected orientation and scale, and whether the result remains stable under small lighting changes. If the detected curve is used for a safety or manufacturing decision, compare with labeled examples and physical measurements. Canny provides useful low-level structure, but downstream logic must decide which edges correspond to the object the user actually cares about.