Medical Image Segmentation Algorithm Based On Fuzzy Set Theory

Image segmentation is a very important research direction in the field of image processing and analysis. The result of segmentation affects whether it is smoothly for the following processing. The fuzziness of medical image is very serious, and fuzzy set theory has good descriptive ability on this kind of fuzziness. So, introducing fuzzy set theory into the field of medical image segmentation can make image have a good segmentation performance.The main content of this thesis are as following:First this paper studies the fuzzy partition entropy method, the method based on fuzzy partition entropy, which is using to transform image segmentation into an nonlinear optimization problem. Then, determine the optimal threshold automatically by the right search strategies. But it is difficult to solve the optimizations. Aiming at this problem we proposed a new method which is to determine parameters by the simple enumeration method. This method uses a conclusion that we can get the maximum fuzzy partition entropy when the fuzzy events have the same probability. The new method can always obtain the optimization solution and have a high accuracy.Then this paper studies the FCM method, this method divides pixels by iterating and optimizing the object function, and it is suitable for to medical image which is uncertain. We need to determine the value of start condition and improve segmentation speed. Aiming at the problem we encounter in the FCM method, we proposed a new method which can determine initial conditions automatically. This method can eliminate the opportunities of getting a local optimum. We also proposed another method that using FCM at edge areas and using K-means at no-edge areas. This method simplified the computation because no-edge areas are important parts of medical image.At last this paper studies the segmentation method based on fuzzy connectedness, this method is a combination of image segmentation, graph theory and fuzzy technology. It is a very complex algorithms, and has a good segmentation performance, but it need to compute the fuzzy connectedness of each signal pixel. Aiming at this problem we proposed a new method. At the beginning we need to pre-process the medical image. Then we have to find connected areas by the spanning tree method. Through this method, we reduced the complexity of algorithmic.