Modeling for Watershed Trasform and Development of Watershed Algorithm
ZHOU Hai-fang, JIANG Yan-huang, YANG Xue-jun (Department of Computer Science and Technology National University of Defense Technology, Changsha 410073)
Watershed transform is a classical method of image segmentation in mathematical morphology. This method, with a wide perspective, has been applied successively into some fields like remote sensing images processing of satellite and radar, biomedical and computer vision applications. Watershed transform is a relatively time consuming task, and the study of this problem has been disesteemed by researchers for its low efficiency. But also for this reason, there appear some new theories and algorithms; moreover, with the development of parallel tools, watershed transform has an increasing attention internationally. As to the resent literatures on this field, variety of formalizing model makes the definition, algorithm and implementation of watershed lack in uniform description. Pointing to this status this paper starts with a rigorous definition of watershed transform for continuous case, followed by two kinds of representative definitions for digital case. The solution of plateau problem is discussed in detail. Then some new watershed algorithms proposed in recent years are classified and analyzed, including watershed algorithms by immersion and by topographical distance. The need to distinguish between definition, algorithm specification and algorithm implementation is pointed out. Finally we present the problems and challenges of future research, including accuracy of watershed transform, problem of over-segmentation, parallelization of algorithms and so on.