Automatic segmentation of diatom images

In this paper we present a segmentation technique based on tools from mathematical morphology which can be successfully used for automatic segmentation of diatom images. The proposed method belongs to the class of hybrid segmentation techniques, and is based on the morphological watershed from markers. The novelty of this method is the computation and selection of markers. A new connected operator is used to simplify the input image and to produce candidate marker regions. A further step which selects among these regions is carried out in order to produce the final markers as a label image, and a watershed process initiated from this image is applied on the gradient of the input image. In a post-processing step, the contours of the diatoms present in the resulting image (given as watershed lines) are extracted.