Machine Vision and Applications (2006) 17:297–307 DOI 10.1007/s00138-006-0035-1 ORIGINAL PAPER A model of diatom shape and texture for analysis, synthesis and identification Y. A. Hicks · D. Marshall · P. L. Rosin · R. R. Martin · D. G. Mann · S. J. M. Droop Received: 16 March 2005 / Accepted: 23 May 2006 / Published online: 11 July 2006 © Springer-Verlag 2006 Abstract We describe tools for automatic identification and classification of diatoms that compare photographs with other photographs and drawings, via a model. Identification of diatoms, i.e. assigning a new specimen to one of the known species, has applications in many disciplines, including ecol- ogy, palaeoecology and forensic science. The model we build represents life cycle and natural variation of both shape and texture over multiple diatom species, derived automatically from photographs and/or drawings. The model can be used to automatically produce drawings of diatoms at any stage of their life cycle development. Similar drawings are tradition- ally used for diatom identification, and encapsulate visually salient diatom features. In this article, we describe the meth- ods used for analysis of photographs and drawings, present our model of diatom shape and texture variation, and finish with results of identification experiments using photographs and drawings as well as a detailed evaluation. Keywords Classification · Automatic drawing synthesis · Principal curves · Texture analysis · Diatoms 1 Introduction Diatoms are unicellular microscopic algae found in prac- tically any moist environment. The study of diatoms is of Y. A. Hicks (B ) Cardiff School of Engineering, Cardiff University, Queen’s Buildings, P.O. Box 925, Cardiff, CF24 0YF Wales, UK e-mail: hicksya@cf.ac.uk D. Marshall · P. L. Rosin · R. R. Martin School of Computer Science, Cardiff University, Cardiff, UK D. G. Mann · S. J. M. Droop Royal Botanic Garden Edinburgh, Edinburgh, UK importance for a variety of reasons, one being that they per- form 20% of all the photosynthetic fixation of carbon diox- ide and hence generate 20% of the oxygen produced each year [8]. Identification of diatoms, i.e. assigning a new spec- imen to one of the previously described species, finds appli- cations in ecology, palaeoecology and forensic science. Each diatom has an outside silica shell; the shell contains two larger elements called valves, one on either side of the cell, which bear species-specific patterns. Many diatom valves are sufficiently flat to give a repeatable two-dimensional (2D) view in all photographs. In the rest of the article, we refer to the external 2D contours of such valves as the shape of the dia- toms, and to the valve patterns as the texture of the diatoms. There is a great variety in the size, shape and texture of such silica shells, and these characteristics are traditionally used by experts to identify new diatom specimens by comparing them to photographs and drawings of previously described species. The identification task is very difficult due to the huge number of species estimated to exist (approximately 2 × 10 5 [7]), close similarity in appearance of some species on one hand, and individual and life cycle related variation in diatom size, shape and pattern within a species on the other hand. Over the years, a wealth of diatom drawings and pho- tographs has been accumulated in the biological literature (Fig. 1). Recently, work has been undertaken to digitise this material and place it in searchable databases (e.g. [15]). A system for automatic identification of diatoms using digitised photographs was a natural progression (e.g. the ADIAC pro- ject described in [1]). However, to the best of our knowledge, there has not been any research done on automatic identifi- cation of diatoms in drawings. Inclusion of digitised drawings in the system and provid- ing the ability to compare photographs and drawings can pro- duce several benefits. First of all, drawings contain mainly