Development of an active contour based algorithm to perform the segmentation of soot agglomerates in uneven illumination TEM imaging

Abstract

Automatic methods for morphological characterisation of nano-particulate from microscopy images usually involve the use of thresholding procedures, which cannot effectively deal with inhomogeneous backgrounds. In this paper, an algorithm based on an active contour model is proposed to extract soot projections from uneven illumination TEM imaging. The presented method involves a pre-processing stage where a median filter and a size normalisation step are performed. Then, the model applied is based on a level set formulation with local region fitting for noisy image segmentation. Once the active model reaches a convergence result, soot regions are isolated and labelled even under inhomogeneous illuminations. The accuracy of the proposed algorithm was evaluated by statistical comparisons of soot size descriptors extracted from 80 testing images, with results obtained by the application of the Otsu's segmentation method and manual measurements. Excellent agreement was found for uneven illumination micrographs without compromising the performance on regular imaging.