Medical Image Segmentation Based On Support Vector Machines

Image segmentation is a widely studied and applied technology in image processing field. Along with the improvement of the actual application requirements, various theories and methods of image segmentation are constantly expanded, and the applications of image segmentation are more extensive. Image segmentation has been widely applied in medical image analysis. Although the current segmentation methods have achieved a certain effect, the previous segmentation techniques are mostly based on the traditional statistical methods, and they are based on the asymptotic theory when the number of the samples tends to infinity. In the problems of high dimensional feature and small sample number, they are so difficult to obtain good results and their generalization abilities are poor. But in practical applications, the number of the samples is often limited, so some very good methods of learning theory in the practical problems may not be satisfactory. Support vector machines which are emerged in recent years, are based on the theory of VC dimension of statistical learning theory and the principle of structural risk minimization. According to the limited information of the samples, support vector machines find the best compromise between model complexity and learning ability in order to obtain the best generalization ability. Traditional statistical learning methods find the optimal value when the number of the samples tends to infinity, while the support vector machines obtain the optimal solution under the limited information of the samples, so their generalization ability is better than the traditional learning methods.In this paper, we mainly do some research on medical image segmentation, using the method of support vector machines to segment blood cells image. Combining the specific analysis of experiments and the statistics of data, we illustrate the effects which the kernel functions of support vector machines, the kernel parameters of kernel functions and the penalty factor C do to the performance of segmentation. After adding noise to the image, the performance of support vector machines has been described. At the same time, we study the impact of the characteristics of input samples on the segmentation results. The result of comparing the performance between support vector machines and other segmentation methods shows the advantage of support vector machines. Through the experiments of different numbers of training samples to the segmentation results, it has validated that the method based on support vector machines is better used in the field where the number of the samples is small. The method of support vector machines is a new method of machine learning, and their characteristics need to be studied further more. The application should be expanded to more areas, and there is more work follow-up to be done.