Research On Medical Image Segmentation Algorithm Based On Regional Level Set Theory

With the development of computer technology,image segmentation,as a key part of image processing,is of great significance in the field of pattern recognition,object tracking,and image understanding.Some scholars have devoted to the research of image segmentation,and put forward many segmentation algorithms.Among them,the image segmentation method based on the level set is promising.The method uses the zero level set of the level set function to indirectly represent the contour curve of the target.It can flexibly handle the complex structural changes such as splitting and merging of the contour curve in the evolution process.Although some level set image segmentation models have been proposed one after another,there are still some problems to be solved.For example,for the image with low signal to noise ration,complex background or intensity inhomogeneity,there are some problems such as low segmentation accuracy,low segmentation efficiency and high algorithm complexity.Based on the level set theory,this dissertation studies the medical image segmentation algorithm,which mainly includes the following aspects:1.A level set segmentation algorithm based on local information entropy is proposed.The gray level uniformity of the image has a high correlation with the corresponding local information entropy value,that is,the greater the gray level change of the region,the greater the corresponding entropy value of information.Firstly,local information entropy is used as a measure of image gray uniformity and only needs to be calculated once during the entire numerical solution process.There is no need to recalculate in each iteration of the level set function,so the complexity of the level set evolution process has not increased.Secondly,the K-means clustering criterion of the LIC model is used to collect the gray information of the original image to correct the bias field of the image.By introducing the regular term of the level set function into the energy functional of the proposed algorithm,we can prevent the level set function from being re-initialized frequently during its evolution.In the proposed algorithm of this dissertation,the combination of the K-means clustering criterion and local information entropy can not only improves the segmentation accuracy and efficiency of image with noise and intensity inhomogeneity,but also has stronger robustness to the initial contour.2.An adaptive level set segmentation algorithm based on energy information is proposed.The advantage of the global region energy fitting term is that it is not sensitive to the initialization of the level set function,and robust to noise.The disadvantage is that it cannot segment the intensity inhomogeneous image.The advantage of local region energy fitting term is that it can segment the intensity inhomogeneous image,and the disadvantage is that it is more sensitive to the initial position of the evolution curve.In the process of segmenting the target regions from the image,the maximum number of iterations that used to control the evolution curve needs to be set manually according to experience,which is tedious,time-consuming and inflexible.Based on the combination of local energy fitting term and global energy fitting term,this dissertation firstly introduces the reaction-diffusion function and edge detection function in the regular term,so that the level set function can maintain an approximate symbol distance function during the curve evolution,thereby avoiding reinitialization and reduction computational complexity.Then,based on the energy change rule of the level set model in the process of image segmentation,the evolution termination criterion of level set function is defined,so that the level set function can adaptively terminate iteration when the target is accurately segmented during the evolution process,avoiding the tedious work of manually setting the maximum number of iterations in the evolution process.Finally,the experimental results show that the proposed algorithm improves the accuracy of image segmentation compared with the traditional algorithms.3.A new level set segmentation algorithm based on local bias is proposed.In practical application,the intensity inhomogeneity of image brings many difficulties to the image segmentation effect and image understanding.In order to solve the segmentation problem of the image with intensity inhomogeneity,by analyzing the illumination model,this dissertation firstly uses the Chebyshev basis function to ensure the smoothness of the bias field,making the estimated bias field more reasonable and accurate.Secondly,for the image with serious intensity inhomogeneity,the local deviation matrix is introduced and the energy functional of local clustering is redefined by considering the deviation of local region between the measured image and the real image.The deviation matrix can be used to correct the bias field with a large change in the local neighborhood in the form of iteration during the curve evolution,and correct the bias field estimation value in time.In this dissertation,the energy functional formulas for segmenting two-phase image and multi-phase image are given respectively,and the corresponding level set function has also been given.By minimizing the energy functional of the proposed model,the segmentation results and the estimation of the bias field can be obtained simultaneously.The experimental results show that the image segmentation accuracy of the proposed model in Brain Web database is higher than that of other models.