[A robust classification method for five types of leukocytes in peripheral blood based on mean-shift clustering].

A new leukocyte classification method for recognition of five types of human peripheral blood smear based on mean-shift clustering is proposed. The key idea of the proposed method is to extract the texture features of leukocytes in a visual manner which can benefit from human eyes. Firstly, some feature points are extracted in a gray leukocyte image by mean-shift. Secondly, these feature points are used as seeds of the region growing to expand feature regions which can express texture in visual mode to a certain extent. Finally, a parameter vector of these regions is extracted as the texture feature. Combing the vector with the geometric features of the leukocyte, the five typical classes of leukocytes can be recognized successfully using artificial neural network (ANN). A total number of 1 310 leukocyte images have been tested and the accurate rate of recognition for neutrophil, eosinophil, basophil, lymphocyte and monocyte are 95.4%, 93.8%, 100%, 93.1% and 92.4%, respectively, which shows the feasibility and high robustness of the proposed method.