Study On The Automatic Segmentation Of Blood Cell Image

The recognition of white blood cell is very important in clinical blood examination,while in many hospitals there exist quite a few problems at present. Take the staff in the hospital for instance, they are likely to make improper decision resulted from fatigue, which may delay diagnosing the patients due to the lower efficiency. Therefore, it is of much significance to produce a system to recognize white blood cells automatically. At the same time, numeralization is the tendency of the development of biology, and image is the major tool of numeralization. So, other studies of biology can borrow ideas from this thesis.The system, which can analyze blood image and classify white blood cell, is theoretically on the grounding of computer, image processing, pattern recognition and artificial intelligence. The principle of the system originates from that of eye to brain. In this thesis, some relevant worldwide research and the actuality of the system are discussed. And what follows is a practical system to classify white blood cells automatically and its image storing structure, which mainly consists of microscope, computer and CCD camera and applies technology of image analysis into practice.On the basis of the system mentioned above, a study is presented on the approach to applying image processing technology into color cells image to achieve automatic segmentation of white blood cells. First, original image is changed from RGB color space to HSI color space, and then saturation image is isolated. Second, the histogram of saturation image is analyzed through multiresolution based on wavelet transform, i.e. the histogram is decomposed by Gaussian functional second derivative as a wavelet function, the threshold is selected on the lowest resolution, and then the histogram is reconstructed by the wavelet function and the threshold is tracked until the highest resolution. Therefore the optimum threshold is acquired and the nucleus of white blood cells are segmented by the optimum threshold segmentation. Thirdly, whole white blood cell is segmented by the method of watershed given that the nucleus region is thought of as the original region, and the convexity in white blood cells as a convergence criterion. Last, the feature extraction about white blood cell is briefly discussed.