We present an image processing system for automatically detecting and classifying cyanobacteria species. The system comprises a light microscope, image capture device, and computer hardware and software. Water samples are viewed at high magnification, and images of algae are transferred to the computer for processing. After algae segmentation, properties such as shape, spectral, and textual features are extracted. Classification is implemented via a hierarchy of separate 2-class classifiers. Following classifier training, leave-one-out classification was used to provide a robust estimation of the real classification error. A total of 7 images from 244 were misclassified, indicating an error rate of approximately 2.9%. Although preliminary, the results suggest that accurate classification of algae specimens can be achieved via automated image processing.
Key words: Image processing, pattern recognition, cyanobacteria, Lake Biwa, algal bloom, texture, hierarchical classification.