The Feature Extraction And Recognition For Microscope Cells

The automatic classification and recognition of microscope cell image is a hot topic of biomedical research. The examination of the corporeal components of the microscopic cells of body fluids is the focus of clinical medicine, therefore, it can not only improve the rate of medical testing, and can reduce the subjective interference by achieving the automatic recognition and analysis of microscopic cells image, besides, the microscopy examination technology becomes more scientific and reliable. The paper realized the shape feature extraction, the color and texture features extraction of the microscope cell images, the feature selection and the classification and recognition capability of targets by combining computer image processing and pattern recognition technology. And the researches were applied to the clinical examination apparatus.Shape feature extraction includes tow processes. Firstly, to set up the semantic model of the various cells types’shape feature extraction through the cell image analysis. Secondly, to build the mathematical model according to the shape semantic model, and then achieved it. A shape feature extraction method was proposed. The one is the moment invariants feature extraction algorithm by fusing the unitary spatial distance distribution histogram and Hu moment invariants shape description algorithm, it used the space structure of the ellipsoid to normalize the cell image, and it achieved a robustness feature, the classification is very effective.The color and texture features extraction is another important processing. We firstly established the semantic model of the kinds of cells image’s color and texture feature extraction, and realized it. The paper proposed a color feature extraction algorithm,which extracted satisfying color features and obtained a better effect of image gray stretch through the adaptive probability sliding window histogram; the algorithm was achieved by taking the regional average gray value as the center value of the sliding window,and make an building an adaptive threshold of the sliding window based on the maximum and minimum of probabilities. Besides, we proposed a multi-fractal dimension feature extraction algorithm which combined the weighted mean value and the spatial information of the grid, it got a effective texture feature, which reflected the gray-scale information and reduced the influence by the count of the cell image area. To complete a better robustness of multi-fractal dimension algorithm, an improve multi-fractal dimension texture extraction algorithm was proposed, which joined the spatial structure of the ellipsoid again and obtained a geometric invariant texture feature.To found a model of the feature selection, the paper presented a improved feature selection algorithm which associated a mean-variance identified function and the PCA. The algorithm got an orthogonal feature space based on the K-L transformation, then projected the original feature space onto the orthogonal feature space, finally, selected the feature vector in accordance with the descending classification capability of these vector by the mean-variance function. It compressed the feature space effectively and accelerated the speed of system working.Microscope cell image pattern classification and recognition was competed after features extraction and selection. The dissertation discussed the design framework of cells image recognition system, furthermore, we simulated and achieved it on the763experimental platform.