Research On The Technology Of Image Auto-Recognition For Bone Marrow Cells

Bone marrow is the main hematopoietic tissue. The analysis on the numbers and shapes of bone marrow cells is helpful to diagnose hematopoietic disease and the other diseases. Traditional method which is being done by human eyes has many disadvantages. The system of image auto-recognition for bone marrow cells which can improve the work efficiency and accuracy for medical workers and overcome the problems of traditional method has a high application value.The main work of the dissertation is contained in 3 parts: (1) studying image segmentation algorithms comprehensively, and locating nucleated cells and its contour correctly; (2) feature selection and feature extraction; (3) classifier for nucleated cells based on BP neural network.The main creative points of the dissertation are embodied in 2 parts: (1) A hybrid strategy based-on watershed transform is proposed and applied to bone marrow cell image segmentation, the strategy can overcome the drawback that traditional marker-based watershed transform is vulnerable to the markers, and the problem that region combine algorithm is slow and inaccurate; (2) Based on co-occurrence matrices and fractals, a new textural feature for color bone marrow cells named co-occurrence matrices and fractals'combining feature (COFF) is proposed, the new textural feature can very well divide Erythron cell and Neutrophil cell and the different moments of the same cell.