On Watershed Segmentation Algorithm Based On Improved Fuzzy Clustering

Image segmentation occupies a very important role in image processing, which widely uses in many fields. Watershed segmentation algorithm based on mathematical morphology is a useful image segmentation algorithm, and has the advantages of rapid, accurate and easy to realize. The watershed algorithm is too accurately, so it leads to over-segmentation problem, not directly using the watershed segmentation result. The method to solve the problem of segmentation:the first method is a pretreatment, which usually uses the watershed algorithm of distance transformation, gradient watershed algorithm and marked watershed algorithm to reduce false minimum; the second method is reprocessing, according to certain principles, which mergers area after watershed segmentation image. This thesis combined the two methods processed the over-segmentation problem. The main research works of this thesis are as follows:(1) Based on the experiment and analysis, marked watershed algorithm reduced false minimum, achieved the purpose to reduce the over-segmentation regions. But seen from the image of watershed segmentation algorithm used marked, the results of still had a lot of over-segmentation regions.(2) In order to solve the over-segmentation of watershed segmentation algorithm, this thesis post-process the image of marked watershed segmentation, which used FCM clustering algorithm of density to merge region of watershed segmentation.The FCM clustering algorithm of density comprehensive considered the distribution of data, to a certain extent, which reduced the over-segmentation phenomenon.(3) Based on type-1fuzzy set theories, the traditional FCM clustering algorithm can not well deal with the uncertain problem in the practical application. Type-2fuzzy set dealt with uncertain aspects ability stronger than type-1fuzzy sets, then the FCM clustering algorithm of type-2fuzzy sets can better deal with uncertain problem. In addition, this thesis applied Voronoi distance on cluster, so the point in the same area was no longer to consider membership degree. This thesis combined the two methods and the fuzzy kernel clustering obtained a new kind of fuzzy kernel clustering algorithm, which is interval type-2fuzzy kernel clustering algorithm of Voronoi distance.(4) This thesis applied interval type-2fuzzy kernel clustering of Voronoi distance on marked watershed algorithm, at the same time, which considered the gray information and spatial information of images. This thesis realized multi-stage region merging. The watershed segmentation algorithm dealt with noise image, which significantly reduced the over-segmentation problem caused by watershed segmentation algorithm.