Segmentation of non-stochastic surfaces based on non-subsampled contourlet transform and mathematical morphologies

In precision engineering, non-stochastic surfaces are employed more and more widely in advanced functional components. The statistically defined amplitude or spatial parameters commonly adopted for stochastic surfaces are not suited to characterize non-stochastic surfaces. It is required to segment the whole surfaces into regions and assess the qualities of the geometrical features individually. The non-subsampled contourlet transform (NSCT), composed of bases oriented along various directions in multiple scales, is a shift-invariant representation with good directional/scale localization. In this paper, by combining NSCT and mathematical morphologies, a novel surface segmentation method is proposed. The multiscale properties of NSCT make this method flexible in extracting salient borderlines between feature regions, and the mathematical morphological operators are employed subsequently to deal with occasional broken filaments or over-segmentation. Experimental results are presented to demonstrate the superiority of the proposed method on the identification and segmentation of various morphological features with complex boundaries.