Unsupervised segmentation algorithm of HRTEM images

We propose a new unsupervised segmentation algorithm for high resolution transmission electron microscopy (HRTEM) images of nanocomposite materials. The proposed algorithm overcomes two major artefacts affecting these images, high level of noise and nonuniform illumination. The identification of the particles is based on the generation of a hierarchy of lower resolution images and a pixel-based segmentation at the highest level. The second part of the algorithm deals with the accurate determination of the border of the particles by means of snakes and a final contour refinement through the hierarchy of images. The experimental results obtained over real HRTEM images show a good performance in terms of both, detection of particles and determination of their borders.