The Segmentation And Feature Extraction Of Marrow Cell Images

Objective and quantitative analysis of microscopic images of cells and tissues has been a goal of human pathology and cytology. The automated, computer-based quantitative analytical techniques provide powerful tools for the study on human morphological structure and physiological activities, and they are objective, quantitative with high speed and precision. In this thesis, thorough research is carried out in methods of segmentation in microscopic images and representation of nuclear texture.The decomposition of complex scene into their component is one of the most difficult problems in quantitative analysis of microscopic images. The segmentation heavily influences the subsequent feature extraction and classification. In this study, the four methods for the segmentation of overlapped cells are presented. The erosion and dilation method, geodesic reconstruct algorithm and a new improved method are based on binary morphology, so their errors of segmentation mostly depend on that of the threshold algorithm they used. The watershed algorithm utilizes the prior knowledge that overlapping part would contain weak gradient. By using the watershed algorithm, better results are obtained. Comparison of these four methods is also made in detail.Texture analysis of nuclei is useful in evaluation of the heterogeneity of complex structure. A set of features are extracted based on the theory of fractal geometry to classify different kinds of cell. Improved methods are also proposed to avoid the boundary effects.The simulation and the experiment shows that the methods proposed in this thesis can achieve satisfactory results for complex microscopic images with different purpose by making use of all kind of feature and information, and that the fractal texture feature can effectively represent cell structures.