Identification of Diatoms by Grid Graph Matching

Abstract

Diatoms are unicellular algae found in water and other places wherever there is humidity and enough light for photo synthesis. In this paper a graph matching based identification approach for the retrieval of diatoms from an image database is presented. The retrieval is based on the matching of labeled grid graphs carrying texture information of the underlying diatom. A grid graph is a regular, rectangular arrangement of nodes overlaid on an image. Each node of the graph is labeled with texture features describing a rectangular sub-region of the object. Properties of gray level co-occurrence matrices as well as of Gabor wavelets are used as texture features. The method has been evaluated on a diatom database holding images of 188 different diatoms belonging to 38 classes. For the identification of these diatoms recognition rates of more than 90 percent were obtained. 1 In terms of biologists diatoms are hierarchically classified in genus, species, subspecies and so forth, but in this paper we'll use the term class in the pattern recognition sense.