Research Of Template Matching Microscopy Cell Image Segmentation Algorithm Based On Clustering

In medical image processing, segmentation of microscopic cell image is one of the most basic and essential step. Especially in the aspect of medical diagnosis and disease detection, it can save a lot of manpower, and provide a relatively objective segmentation result. Depending on cell’s feature of gray, shape and texture, microscopic cell image segmentation is the process of extracting the cells from Microscopic cell image. So far, a large number of microscopic cell image segmentation methods have been proposed. This paper reviews several main microscopic cell image segmentation methods’ ideas and points out their advantages and disadvantages. Experiments show that, microscopic cell image segmentation has the questions of the cell characteristics of cell image, sample production and cell boundary fuzzy. In accordance with those questions, several common microscopic cell segmentation methods have poor performance. Their segmentation results include false edge significantly, empty holes in the internal cells region, and over segmentation phenomenon. At the same time, the common microscopic cell image segmentation methods are usually suitable one type of microscopic cell image segmentation, which are not universal for many types of microscopic cell image segmentation. So we draw forth the algorithm framework of template matching method, which not only can get good segmentation results, but also can be more fitting to cell’s edge. But the algorithm framework of template matching method has a high error rate, when dealing with the image whose cell and background has low contrast and whose cell morphology has a big difference. Hence we need to improve this method. We regard the research of the template matching method’s algorithm framework as a point, and put forward two main innovation works.①We put forward a kind of microscopic cell image template matching segmentation method based on Clustering. At first, by wavelet invariant moments, we extract the shape feature of the train samples for clustering and statistic the main variety of shape and texture respectively. So it can create a series of templates. At last, by the maximizing normalized cross correlation, we find the best matching location between the cell and the template in microscopy cell image, so as to segment the test image. The contrast with other methods shows that, the method for microscopic cell image segmentation has a certain universal. It can be fit for the microscopic cell images, which have small differences in cell morphology and simple background. At the same time, it is also fit for the microscopic cell images, which have a complex background and big differences in cell’s shape.②According to the characteristics of microscopic cell image, we put forward a new method for clustering validity’s evaluation, which is combined with clustering algorithm. So we realize a template matching segmentation algorithm based on adaptive clustering. This segmentation method can automatically select the optimal clustering number. The experimental results show that, in the premise of not affecting the segmentation performance, the method realizes the function of automatically determining the number of clustering.