The Research On Marrow Cell Image Segmentation Methods Production Process

In the process of automatic identification of cell images, cell segmentation is one of the most important steps. Segmentation results have a direct relation to the following steps of cell images identification. Because of the diversity and complexity of the cell images, at present, there is no method can correctly separate all the cell images. No matter which segmentation algorithm is selected, it will always have overlapping cells in the cell images. So finding a new segmentation algorithm which can resolve the problem is urgent.The image segmentation of leukemia cells in bone marrow smear is studied in this paper. The major contributions are as follows:This paper described research status in this field, introduced the region-based segmentation, edge based segmentation and segmentation based on a specific theory of several commonly used segmentation algorithm and a comparative analysis, and pointed out their advantages and disadvantages.In the bone marrow image preprocessing, in order to match the complexity of the leukemia cells in the block image background with noises, an image preprocessing method that first uses bilateral filtering method to determine the threshold via the OTSU method is proposed, instead of using the traditional image preprocessing using median filters.In the bone marrow cell image segmentation process, we proposed energy-based non-uniform projection of the image block, use wavelet to extract image feature, and developed a modified FCM algorithm to cluster feature vectors.In resolving the issue of adhesion of cells, an important characteristic of the algorithm is that it passed the domain pixel information to calculate pixel minimum distance from background pixel dot pitch to recently shortest distance between points, used a linked list structure to store the pixel point from the update information, and uses a matrix iteration to achieve distance transform. Also, the mathematical morphology and watershed algorithms are used at the same time from all the core areas, fast recovery of the original cell area, and find the exact separation point.Finally, we provide a summary of the paper, and point out future work that will focus on narrow goals and objectives of irregular single-particle image separation of polyethylene reactor targets.