Component Tree Based Algorithms For Medical Image Segmentation

With the rapid development of computer science,medical imaging technology has also been greatly improved.As the foundation of medical imaging technology,image segmentation has become one of the research hotspots in that field.Because of the complexity of medical images,medical image segmentation is a difficult problem in image segmentation.In recent years,scholars have proposed a large number of excellent medical image segmentation algorithms,such as pixel based algorithm,region based algorithm,model based algorithm,and artificial neural network segmentation algorithm,but all these algorithms are ideal for specific type of image segmentation.Component tree was firstly developed in the field of statistics.Later it was incorporated into the theoretical framework of mathematical morphology,and became a morphological method.In this paper,a medical image segmentation algorithm based on component tree is studied.The feature of this algorithm is to integrate the pixel information and the region information of the image.After building the image component tree,the image processing is directly transformed into the component tree processing,which provides a new way of image processing.The main work of this paper is as follows:Because noise is unavoidable in image,filter operation is first done before building component tree for image.In the preprocessing stage,we use Gauss-Laplasse(LOG)to do image filtering.The image after LOG filtering keeps the details while suppressing the noise.After constructing the component tree for the image,the difference of the gray level between the adjacent layers is not obvious.It is possible that the nodes between the adjacent layers belong to the same target area.The size of the component tree is different between the different nodes of the same layer and the size of the area is different,because the image has noise and the possible node area is too small.Aiming at the shortcomings of the above component tree,we need to reduce the nodes,so we propose a criteria for attribute similarity merging of adjacent components.According to this rule,merging components tree nodes to achieve the purpose of image segmentation.Other scholars basically use one or more attributes of component tree nodes independently.The process modules are combined in this paper,the integrated use of node's two attributes area and height.From the experimental results,we can see that the image segmentation algorithm based on the set tree proposed in this paper improves the segmentation effect of the target.