A method for automatic identification of diatoms (single-celled algae with silica shells) based on extraction of features on the contour of the cells by multi-scale mathematical morphology is presented. After extracting the contour of the cell, it is smoothed adaptively, encoded using Freeman chain code, and converted into a curvature representation which is invariant under translation and scale change. A curvature scale space is built from these data, and the most important features are

more »

... ed from it by unsupervised cluster analysis. The resulting pattern vectors, which are also rotation-invariant, provide the input for automatic identification of diatoms by decision trees and k-nearest neighbour classifiers. The method is tested on two large sets of diatom images. The techniques used are applicable to other shapes besides diatoms.