Diatom Screening and Classification by Shape Analysis

Abstract

Nowadays diatoms, microscopic algae, constitute a favorite tool of modern ecological and evolutionary researchers. This paper presents a new method for the classification and screening of diatoms in images taken from water samples. The technique can be split into three main stages: segmentation, object feature extraction and classification. The first one consists of two modified thresholding and contour tracing techniques in order to detect the greater amount of objects. In the second stage several features of the segmented objects are extracted and analyzed. The last stage calculates the significant centroids, means and variances, from the training diatom set and then classifies the main four diatom shapes founded according with the Mahalanobis distance. The samples are normally contaminated with debris, that is, all particles which are not diatoms like dust, spots or grime; which makes necessary to selected a threshold value for rejecting them. The results show the method ability to select 96% of the used diatoms.