An Improved Iterative Watershed Algorithm For Image Segmentation Based On Ridge Detection

With the rapid development of automatization in metallurgical process, imageprocessing and segmentation technology become more and more indispensable andessential. Since metallographic images of metal crystal structures have disconnectedboundaries and heterogeneous intensity, the traditional watershed segmentation algorithm,comparing to other image segmentation algorithms, has better results. However, duringtraditional watershed image segmentation, false combination of "pseudo-blob" caused by"over-segmentation" debases the accuracy of segmentation.This paper, based on traditional watershed segmentation algorithm, firstly presents anew concept called "real-blob"."Real-blob", as opposed to "pseudo-blob", refers to thecomplete blob which satisfies the "real-blob" classification rule during combination of"pseudo-blob" and it consequently is circumscribed to merge surrounding blobs.According to this new concept, this paper also presents a "real-blob" classification ruleand mathematical formula as opposed to "pseudo-blob" classification rule, andsubsequently modifies the "pseudo-blob" classification mode and "pseudo-blob" mergerule. Thus the improved algorithm not only can correctly remove the "pseudo-blob", butclassify the "real-blobs" instantaneously. Since the "real-blobs" can no longer falselymerge other pseudo-blobs during the process of pseudo-blobs combination, the finalresult could be more precise.The result shows that this new image segmentation algorithm not only avoidswrong merging, but also maximizes the number of the pseudo-blobs elimination whichgreatly improves the accuracy of the segmentation. Simultaneously, the stability andreliability of this revised software have been considerably ameliorated.