Image Processing for Automated Identification of Harmful Algae

The pre-processing, segmentation and feature extraction of harmful algae from micro-image are presented. Theoretical analysis and experiment results show that 2D entropic threshold algorithm is capable of segmenting the harmful algae’s image effectively, as compared with other auto-segmentation methods. Mathematical morphology was applied to increase the veracity of feature extraction. Feature extraction and classification are accomplished by using the image edge and texture. Experiment result indicates that the classification accuracy is over 80％, as compared with the artificial identification.