A Scanning Electron Microscope image segmentation method for steam generator fouling rate estimation

In this paper, we propose a new approach to segment Scanning Electron Microscope (SEM) images and separate particles from background. The automatic segmentation of SEM images is a challenging problem owing to illumination non-homogeneities and image variability. We take advantage of priors on the inspected material as well as complementary clues (intensity, space and contours) to get a segmentation method as robust as possible on a wide range of images. The proposed algorithm first applies the mean-shift clustering procedure, which consists in successively detecting the modes of the image probability density function and merging them with intensity and contour clues. Then a spatially constrained K-Means algorithm alternatively estimates the bias field and classifies pixels between crystal and background. This method was successfully applied to compute the fouling rate of a nuclear plant steam generator tube mock-up.