Research On Ferrography Images Segmentation Based On Mathematical Morphology

At present, the wear of mechanical parts is the main reason of the mechanical faults. So it’s very significant to make an accurate analysis of a machinery’s worm state without disassembly and turning off . Analysis of wear particle is the key in the research on ferrography image to check the machine’s severity and running status by analyzing the wear particle’s type, composition, concentration and size. The ferrography image segmentation play a decisive role in the precondition to analyze the wear particle and provide the basis for the subsequent analysis of the wear particle image.Firstly, Considering on the advantage of resisting noise performance, operation efficiency and accurate positioning, mathematical morphology method is used to preprocess ferrography image which proposed in this paper. In order to distinguish the ferrographic image’s non-uniform background, the technology that based on using morphological geodesic dilation is proposed, it use morphological close-open filter to filter noise, and use dilation-erosion to fill with particle of the pores. Base on these, the ferrography image avoid for over-segmentation. In addition, by using morphological edge detection, a continuous and closed wear particle edge will be got.Secondly, this paper he improved watershed segmentation method is mentioned to segment the preprocessed ferrography image. Morphology erosion-ultimate is used to obtained the minima regional at first, and use dilation conditional to segment particles. To avoid over-segmentation , searching the markers is improved in this paper. In this paper, the distance transform technique based on the serial algorithm is proposed to construct the topology image, with the searching the markers with the local maximum line method and merging the duplicate tags in the same region. The two methods of the organic combination may not only effectively segment the particles of ferrographic image, but also may be a good solution to solve the over-segmentation of the traditional watershed segmentation method caused by the noise sensitive.Finally, by experiment and comparison with other algorithms, this paper proved that the algorithm above can effectively solve the image processing problem like noise suppression, edge detection, image segmentation and so on.