Research On Active Contour Mumford-Shah Model Based On Local Feature Vector On CT Image Segmentation

With the development of computer vision,virtual reality,artificial intelligence and other technologies,image processing becomes more and more important in computer research.Considering the fact that image segmentation is the link between extracting image data and image analysis,scientists are always focusing on image segmentation research.Several fundamental algorithms and almost hundreds of relative algorithms have been put forward since image processing became one of the key domains in computer science.Snake and Mumford-Shah,as the two main models of active contour models and curve evolution,have been studied intensely in nearly three decades.We put forward a improved algorithm based on standard Mumford-Shah model in this paper.Basically,we improve the model in two aspects after analyzing pertinent algorithms,one is the initialization strategy,the other is the minimization of the energy functional.As for the initialization strategy of Mumford-Shah,we use PCA to reduce the dimensionality of images,and then we use K-means clustering to figure out the initial position of the segmentation curve.After the preprocesses of image data,we use Mumford-Shah model to have the image split.Using standard MumfordShah model to find out the optimal segmentation is equivalent to solving the minimum of energy functional,therefore,we need to figure out the solution of calculating the energy functional.We use one of the convex relaxation techniques to optimal the discrete problem of energy functional,and then use the Chambolle-Pock dual algorithm to calculating the extremum,so that we can get the optimal image segmentation.Since the improved algorithm we put forward is mainly for medical CT images,and we are focusing on spleens especially,we use local spectral histogram algor ithm as our strategy of extracting feature from images.We use this improved algorithm on medial CT images in our experiment,and to make the conclusion more objective and persuasive,we also experiment these images on other similar and relevant algorithms.After contrast experiments,we can get the conclusion that this improved algorithm based on standard Mumford-Shah model has a better performance than some other image segmentation algorithms in CT images,furthermore it has a strong robustness.