Abstract: At present, the microscopic image identification of harmful algae based on biological morphological features are applied widely. However, there are some challenges, such as limited algae species resources, the difficulty of describing biological detailed features. To solve these problems, the identification of harmful algae was divided into four gradual processes: precise segmentation, feature extraction, feature dimension reduction and classification. First, the method based on projection and integral on multiple directions was proposed to extract cell object, and then the spine and cingulum were extracted. Second, the effective description were proposed after shape features, moment invariants, texture features and domain specific features were extracted. Then, SVM classifier was designed to recognize objects. Experimental results showed that this algorithm could achieve the quicker and better segmentation and recognition of 15 species of algae.

Key words: cingulum segmentation, microscopic image recognition, spine segmentation, biological morphological features