Study On Method Of Automatic Segmentation And Recognition Of Blood Cell Images

It is an important topic for automatic segmentation and recognition of blood cells in the field of medicine image research based on computer image processing and pattern recognition technique. At present, a lot of small and medium-sized hospitals still use manual visual methods to recognize and count blood cells. Some automatic blood cell analyzers used by hospitals can only be carried out to recognize three kinds of white cells, so that they can not completely meet the clinical needs. Therefore, it is an important direction of development that using digital image processing approaches and pattern recognition to segment white cells and classify them.Through looking up some information about blood cells images and doing experiments, we find that the gray values of the G component and S component of white cell nucleus and other regions in microscopic images of blood cells have a threshold. We improve the MEANSHIFT algorithm, the G-channel image histogram and S component histogram of the blood cell images are processed by using a method based on improved MEANSHIFT algorithm, then we get the adaptive thresholds that can segment white cells and other regions. Then, the white cell nucleus is separated from the image through the threshold segmentation. We identify the white cell nucleus based on the method of regional filling and regional growth. And we use the saturation image as the growing rule of the region growing algorithm to get the cytoplasm of the leukocyte. After the feature extraction of white cells regions, we adopt a support vector machine to classify them. Through the experimental comparison, we select the Gaussian radial basis function (RBF) as the kernel function, set the appropriate parameters and construct a two-level SVM classifier. Finally, we get the classification results of blood cells images. The experimental results show that proposed algorithm is robust and adaptive.