Level Set Image Segmentation Model Based On Variable Weights

Image segmentation is the premise of image understanding and recognition,which is the foundation of image processing and the difficult point in the research of image processing and computer vision.In view of the ability to obtain a close and smooth segmentation curve to deal with the topology transformation.Since the model was put forward,many experts and scholars have paid close attention to it,and are still active in the field of image segmentation.After nearly twenty years of development,the theory of level set image segmentation has been gradually improved,and the corresponding segmentation methods have been produced for different application scenarios,such as global information based segmentation model and local area-based segmentation model(LRAC),and so on.However,the research on the combination of image global information and local information is not perfect.The general level set model is used to deal with the kind of image information fusion,which is usually a simple addition of energy Functionals defined by different image information,and a fixed constant weight or a weight that varies with the number of iterations before each item.In fact,in different places in the image,the proportion of the same energy should be different.In this paper,a new algorithm of image segmentation based on the combination of the local region and global region was proposed to solve the dependence of the initial contour on localizing region-based active contour(LRAC)model.The algorithm combines the characteristics of Chan-Vese model and LRAC model.When constructing the level set function,the variable weight parameters were defined to combine the local and global energy functional of the level set function.Different from the way of combining energy term of general level set function,the weight parameters were defined by image gradient and the mean of the inner and outer pixels at the local image.In this way,the local energy is the main component when the target contour is close to the target,and the global energy is the main component when the target contour is far away from the target.In addition,the narrow band method was used in the evolution of the level set function to reduce the complexity of computation time.Finally,in order to verify the effectiveness of the model,a lot of segmentation experiments were done on various types of images using this model,and compared with many existing level set models.Experimental results show that our model has the advantages of both CV model and LRAC model.Compared with LRAC model,the method we proposed relies much less on the initial contour and has a better convergence rate.While compared with CV model,the precision of our model is higher in the effect of target edge segmentation.