Genetic Clustering Algorithm And Its Application In Medical Image Segmentation

Image segmentation is the key step from image processing to image analysis. One of its most important applications is medical image segmentation. Some biomedical information is shown in the form of image, such as X-ray image, fault CT image and ultrasonic image, which makes human vision extend from exterior to interior. By them people can get some useful information of human body's internals in anatomical form, biochemistry and physiological function. Because the focus of medical image is similar to surrounding normal texture in gray and shape so that it is difficult to distinguish with the naked eye, we need to carry out image segmentation so that the focus can show up obviously.Genetic Algorithms is a random, parallel and adaptive algorithm for search and it simulates natural selection and the evolutionary process. It is built by John Holland, the professor of Michigan University of U.S.A in the 70's of the 20^th century. Because its ideas are easy and it is easy to calculate, for recent years it is widely applied in many areas, such as function optimization, combination optimization, automatic control, intelligent control, image processing and pattern recognition, artificial life, machine learning and so on, which make encouraging achievements. Cluster-based analysis is an unsupervised learning approach. It can dig correlative rules from feature data of object of study. As a result, it is a powerful approach to information processing. By combining Genetic Algorithms with cluster-based analysis, this paper proposes a segmentation algorithm based on genetic clustering, which carries out pixels' clustering in feature space. This algorithm applies GA to search optimal clustering center, which avoids the selection of initial clustering center in general algorithm such as C-mean clustering algorithm. Taking a picture of eyeball for example, this paper achieves image segmentation by VC++ programming. This algorithm can get very good effect of segmentation and make the defect show up obviously.