Research On Weighted Chan-Vese Image Segmentation Method

With the rapid development of artificial intelligence and big data technology,image segmentation has been widely used in remote sensing images,medical images and face recognition.However,image degradation caused by factors such as noise,blur,and uneven gray-scale is still a huge challenge in the field of image segmentation.Traditional segmentation techniques include threshold-based and region-based segmentation methods.Compared with the thresholdbased segmentation method,the region-based method considers spatial information,so it has higher accuracy in segmentation tasks,but also higher computational complexity.On the other hand,a large number of manually labeled data is not easy to obtain in clinical practice,so the few-shot image segmentation has attracted attention,but there is also the problem of insufficient segmentation accuracy.The Chan-Vese model is a classic region-based segmentation model,which has the advantages of easy handling of topology changes,high calculation accuracy,and suitable for complex scenarios.Therefore,this paper mainly conducts the following research work based on the Chan-Vese model:(1)A weighted Chan-Vese image segmentation model is proposed.Aiming at the problem of high computational complexity of existing models,a weighted Chan-Vese image segmentation model is developed.The model introduces heat kernel convolution to approximate the replacement of total variation terms,and in order to ensure segmentation accuracy,an adaptive weighted function and higher-order total variation function are introduced.When solving the model,the alternating direction method of multipliers is used to split the model into multiple subproblems that are easier to solve,which improves computational efficiency.In addition,the model is tested in two-phase segmentation and multi-phase segmentation tasks,respectively.The results show that the segmentation accuracy of the proposed model is better than that of other comparison models.(2)Proposing a few-shot segmentation model combined with deep learning.Aiming at the problem that the traditional segmentation model needs to manually set the initial contour and the accuracy of the deep segmentation model based on few-shot samples,a weighted Chan-Vese segmentation method combined with deep learning is proposed.First,based on a small amount of data,deep learning is used to obtain location prior information.The prior information is then input as the initial contour into the weighted Chan-Vese segmentation model,which improves the segmentation accuracy of the deep network in the case of small samples.Finally,the results are tested on multiple clinical CT datasets and show that the location prior information extracted by the new method effectively improved the segmentation accuracy.