Abstract: The segmentation of well-contrasted objects is noproblem and the success of several algorithms is provedby applications. But if the objects are poorlycontrasted, it is difficult to find a threshold, whichleads to a right object segmentation and, in manycases, (e.g., touching or overlapping objects) athreshold for the right segmentation of the image intoisolated object regions does not exist. Some methodsare presented which can help to overcome theseproblems. Global information and a priori knowledge areused for the selection of an optimum segmentationthreshold (a threshold is selected independently foreach object). An algorithm for the separation ofconglomerates of convex objects is presented based oncontour information (information about the shape of theobjects). The main characteristics of this algorithmare: construction of a recursive convexity polygon,determination of fuzzy features for the description ofpossible parts of the conglomerate, and dynamicprogramming. Several applications demonstrate the useof further information about shape, grey valuedistribution, and topology. !8
展开▼