Research Of Automatic Recognition System Of Blood Cell Image

It is an important topic in the field of medicine image to make research on Blood Cell image recognition with Computer Pattern Recognition technique. However, there are so many types of blood cells and the image is complex. Therefore, it is not easy to fulfill image recognition with this technique. Up to now, the recognition of blood cells in clinic is by manual work.The system in this paper can recognize the blood cells by pattern recognition technique automatically. The first step is image segmentation. Transform a colorized image(the original image) into a gray image(then second image) and then transform then second one into a distance transformation image(the third image). Identify the regional maximum and dam liner between cells in according to the third image, and select White Blood Cell(WBC) based on the size of cell and gray value, then identify Red Blood Cell(RBC) around WBC with region increase method, and eventually combine the borderline grads of WBC, distance image, identification of RBC with other information to confirms the final profile of WBC. The second step is to select the features of cells. The main features include the followings: radius of cell, the number of nuclear ramous, nuclear - plasma area radio, nuclear gray value, cell-plasma color value, nuclear color value, the roundness value of nuclear, concave value of nuclear, the partiality ratio of nuclear, the texture of cell-plasma, the texture of nuclear. The third step is to design classifier. The classifier is the statistical classification. According to the expert system rules to recognize cells, transform the features of cell into the conditions of judgment, the threshold of conditions arise from the classifier training samples. The experimental results show that the classification accuracy is 96% from the 88 blood cells found in a set of 50 images. The system also confirms the validity and the clinical value.