Research In Segmentation Algorithm Of Peripheral Blood Image

As we know, the number and quality of leukocytes (white blood cells) in human peripheral are important parameters for characterization of human health. Therefore, the statistics and classification of white blood cell are important parts of clinical disease. The automated hematology analyzer is now gradually replacing the traditional manual microscopic counting method because of its simple and convenient. But most of the automated hematology analyzers in domestic market still use electrical impedance analyzer counting method. The accuracy of this method is low. In recent years, in order to improve the recognition rate of white blood cells(WBC), digital image processing technology is widely used in the analysis of white blood cell morphology.The main content of this paper is the technology of digital image processing, applying to leukocyte microscope image. WBC microscopic image preprocessing technology, and the segmentation of WBC are included.In the research of image preprocessing, this paper compares the WBC image features under different color spaces, in order to choose the most suitable color space for cell division. At the same time, some pre-reduction technologies such as image clarity, resolution are analyzed to make the input images are more suitable for the segmentation.The WBC segmentation process is a simulation of the human visual significantly, based on three prior conditions as follow: WBC are nucleated, round and there may be burrs around them, but the burrs will not be greater than the root. According to the recognition of human's eyes, this paper will divide the segmentation process of WBC into three steps: location, region-dependent segmentation and adhesion cell division. We make a deep analysis on the technical difficulties of various segmentation algorithms from the theoretical and practical point of view, and many innovative achievements have been made:1. According to human visual attention mechanism significantly, this paper propose a sequential process: lower resolution image first, then location, the final segmentation. This method not only reduce the segmentation error rate, but also greatly reduce the computation time.2. Improve the traditional watershed segmentation. The error-segmentation and over-segmentation in traditional watershed will be eliminated by compute the grade of membership that sub-blocks in watershed belong to each cell center, and redivide the sub-block produced by over-segmentation.