Watershed Algorithm

The watershed algorithm is an image segmentation technique that treats a grayscale image as a topographic surface, with each pixel's brightness standing in for its height, and finds the boundaries that separate the surface into distinct catchment basins. It works by an analogy to flooding: water sources are placed at the low points, the regional minima of brightness, and the whole surface is gradually flooded from those points; wherever the rising water from two different sources would otherwise meet, a dividing line, or watershed, is drawn to keep them apart, and the resulting set of basins becomes the image's segments. Serge Beucher and Christian Lantuejoul introduced the concept in 1979, and Fernand Meyer developed one of the most widely used efficient implementations in the early 1990s. Run directly on a noisy or highly textured image the technique tends to produce far too many small basins, so most practical uses first smooth the image or seed the flooding at chosen markers before running the algorithm, then merge or filter the resulting regions afterward.