Research Of Level Set Segmentation Models And Their Applications In Medical Images

Image segmentation plays an important role in the process of digital image processing,which lays a foundation for the following image analysis.And in image processing,image segmentation has always been a classic problem due to the problems of intensity inhomogeneity,noise and weak edge.Compared with the traditional active contour model,the main advantage of the level set active contour model is its ability to deal with the complex change of topological structure,and the simple numerical calculation.Because of the local volume effect,overlapping of human tissues or organs in medical images,there has been no uniform and effective method to segment medical images.The application of level set method to medical image segmentation is the classic application of level set model.A hybrid level set segmentation model based on local entropy is devised to deal with the problem of intensity inhomogeneity and initial contour sensitivity which consist in local region-based level set models.Firstly,the local entropy information is added to the LIC model to better describe the local intensity fitting.At the same time,the local CV term is introduced to further improve the availability of local intensity information,so it is preferable to process intensity inhomogeneity.Experimental results show that the model devised in this thesis can effectively segment the image with intensity inhomogeneity and improve the robustness of the initial curve.An adaptive distance regularized level set model conbined with structure tensor is devised to deal with the problem of edge-based DRLSE1 model.The first problem is that the movement direction of the evolution curve depend on the positivity and negativity of the manual set constant coefficient v.The second problem is that the initial curve is fixed inside the target,outside the target or completely surrounding the target,and the segmentation results are dependent on the initial curve seriously.The last one is that the fixed constant coefficient v cannot be adjusted with the image information,making it impossible to detect weak edges and multiple targets.Because a smoothing divergence field v(I)can change the positivity and negativity according to the image gradient information,so the model devised in this thesis introduces v(I)as weight to drive the evolution curve to move inward or outward adaptively according to the image information instead of the original constant coefficient v,and consequently improving the robustness to the initial curve.Also the weight coefficient v(I)contains richer local structure information,it is better to extract weak edges and multi-targets.Experimental results show that the model devised in this thesis improves the arbitrariness of initial curve and detects weak edges and multi-targets more effectively.