Fast Image Segmentation Models Without Initial Contour

Image segmentation is a basic computer vision technique whose purpose is to separate the target object from the image while obtaining the corresponding edge.Image segmentation is widely used in medicine,military,industry and other fields.For a long time,many scholars have conducted a lot of research on how to quickly and accurately segment the parts that people need.In recent years,image segmentation based on partial differential equations has become a research hot spot in image segmentation,where geometric active contour model can be seen successful methods.This dissertation mainly discusses the geometric active contour models and makes appropriate improvements to the existing shortcomings of the existing geometric active contour model.There are two main tasks:(1)Most existing geometric active contour models need to define an initial position for the evolution curve,which easily leads to the image segmentation result being affected by the position of initial contour.This may cause the segmentation results to be affected by the position of initial contour.To this end,a new signed pressure force function is constructed based on the local and global information of the image,and a fast image segmentation model in the form of partial differential equation is proposed.The proposed model is simple and the algorithm process is easy to implement.Experiments show that the proposed model allows constant initialization.It can quickly segment images with piecewise constant intensities,images with intensity inhomogeneity,images with gradually varied intensity,deep images without initial contour.(2)Since the model proposed in(1)only combines the global information of the image,it is impossible to segment images with intensity inhomogeneity.In order to solve the problem,incorporating local information on the basis of(1),a new image segmentation model based on partial differential equations is proposed.The experimental results show that the proposed model can segment images with intensity inhomogeneity without initial contour,and needs only one iteration.