Active Contour Model For Medical Image Segmentation

The goal of image processing is to convert original image into digital symbolic form for better representation and assessment or search and mining image data. Image segmentation is the most basic and important issue in image processing which is to divide the image into multiple regions with similar characteristics based on the given criterion such as image intensity、colok texture and shape, and then extract the object region of interest. With the development of medical image technology, image segmentation has been a critical technology of Computer Aided Diagnosis System. For medical image, the segmentation process consists of two parts, one is to identify the target region in the image such as organ and human tissues, the other is to extract and represent the target region completely. The complexity and diversity of medical images makes traditional segment methods difficult to get a satisfactory result, which are only based on underlying information, As a result, Medical image segmentation has been a bottleneck of CAD system.The image segmentation of active contour based on Level set not only can combine the underlying image information with high level prior information, but also overcome inherent weakness of traditional Snake model, It has widens the utilization field of active contour in dealing with complex images. Traditional active contours always have slow velocity and can’t get ideal segmentation in images with complicated background and been occluded. This paper proposes some improvements on traditional active contours, the results show good performance in medical images and complex nature images. Specific work is as follow:1. On account of drawbacks of traditional C-V model based on regions, we propose an improved C-V model which adds a penalty term and a weight term in original energy function, and applied it in the segmentation of Brain MRI images.2. In this paper, a new Shape-prior based Hybrid Active Contour (SHAC) model is presented for segmentation. By using level set method, this model combines boundary and adaptive region information together and learns an optimal prior shape from the training set. It takes the boundary and adaptive region feature as local information while prior shape as global information. The model combines global and local information in the process of iteration to guide the evolution of deformative curve and achieve the goal of segment target objects. Experiments show that compared with GAC, C-V, and RSF models, SHAC model displays its advantages not only in the segmentation of image strong noise and weak boundary, but also in the image with low contrast resolution and complicated background and contributes improved accuracy.