Image segmentation using Watershed Transform: Marker definition based on fuzzy logic.

The Watershed Transform (W.T.) is a useful morphological tool that allows to distinguish complex structures that can not be processed with image processing conventional algoriths. The W.T. floods the image gradient topography from its local minima. These minima indicates the zones where the flooding starts in order to segment the image. The final result is a labeled image where each pixel belongs to a unique region. Gradient of textured images have irrelevant local minima with low contrast. This results in image sobresegmentation in most of the cases. To avoid sobresegmentation, unique object markers are defined. Markers definition, when objects have different texture, size or shape, requires complex solution highly dependent on each particular application. This paper propose an automatic marker definition method for the Watershed Transform using a Mandami fuzzy inference system. The proposed method is simple, robust and easily adaptable to diferent type of images.