Research Of Leukocyte Segmentation Based On Watershed Algorithm

Nowadays, more and more information technology has been applied to daily life, which also reflected in medical technology. In the medical examination, counts for different types of blood leukocyte are one of the indicators of personal checks. The main processes of automatic cell counter include microscopic image segmentation, extraction and classification, which different with red blood cells and other non-nuclear blood cells. Due to the presence of nucleus in white blood cell, the color tends to be the deepest region of the nucleus in the entire image after dyeing. Further, different types of leukocytes stained a slightly different color. The existence of the nucleus played the role of interference in the conventional method of leukocyte morphological segmentation. A Relatively complete extraction of the nucleus and cytoplasm area is the basis of leukocyte classification. Therefore, segmentation and extraction of WBC region has become the focus of the study.Watershed algorithm has a good response to images containing weak edges, thus making the image segmentation access to more applications. What’s more, watershed algorithm has demonstrated relatively excellent results for segmentation of cell adhesion. But in the microscopic image of white blood cells, due to the presence of nuclei, conventional watershed algorithm based on distance transform, gradient-based and tag-based has more or less problems and difficulties when applied to image segmentation. And because there is the case of white blood cells and red blood cell adhesion simultaneously or instability problems due to staining and image sampling generated, and the diversity of forms in white blood cells, all above brings difficulties to white blood cells for watershed algorithm in image segmentation.This essay firstly analyzed and studied morphological processing algorithms and watershed algorithm, and on this, we propose a new distance-based watershed transform segmentation algorithm to simplify the algorithm. Algorithm larger resolution image of leukocytes, using images obtained by EM clustering replace the nucleus area of the leukocyte cell adhesion segmentation into segmentation. After the replacement of the nuclear region of the image based on the application of distance transform watershed segmentation using prior knowledge of the location of the nuclei of leukocytes, white blood cells to determine the location. Meanwhile, after the distance transformation of morphological image processing, reducing the cell segmentation watershed algorithm is easy to produce the phenomenon of over-segmentation. And because the white blood cell nuclei occupy a large area, condition and easy to obtain, through the watershed can be divided regions merge, thereby further reducing the impact of over-segmentation. This method leukocytes microscopic image segmentation problem into cell adhesion segmentation, simplifying ideas leukocyte microscopic image processing. At the same time proved, segmentation method proposed in this paper on the segmentation accuracy also has a relatively good performance.