Research On Local Region Level Set Method Based On K-means++ Algorithm

In today's social life,everyone is indispensable to obtain image information.Images are also the medium for communication between people and machines Therefore,it is essential to obtain image information and analyze and process the image.Image segmentation is the most basic and key research in the field of image processing,and is the prerequisite for subsequent image recognition and visual analysis.In recent years,the image segmentation technology based on the level set method has received continuous attention and research from more and more scholars.The level set method has a rigorous mathematical theoretical background,and the segmentation result obtained is a smooth and closed target boundary curve with high segmentation accuracy.In addition,this method uses a zero-level set of higher-dimensional functions to express the evolution curve,which can well solve the change of topological structure in curve evolution.There are many excellent level set methods that can well deal with the gray unevenness that is easy to appear in the image,but there are still some problems to be solved,such as the slow evolution of the level set function causes the overall model efficiency is not high;The initial contour model is sensitive and so on.This dissertation starts from an important branch of the level set method:the level set method based on the local region,and proposes three kinds of optimized active contour models based on the local region:(1)An active contour model based on K-means++clustering algorithm is proposed The core of this model is two fitting functions,which are clustered by the K-means++clustering algorithm in a local area window before the evolution of the level set function In this model,a square local area window is first established,and two fitting functions are used to represent the center points of the lighter and darker sub-areas in the moving square local area window.Compared with the traditional region-based active contour model,this method avoids repeated updating of the fitting function during the curve evolution process.Therefore,the proposed model has lower computational cost and can obtain correct segmentation results in less time and iterations.At the same time,the proposed model can effectively segment grayscale uneven images.In addition,experiments prove that the model is robust to the initial contour.(2)An active contour model combining local edge fitting(LEF)energy and RSF(Region-scalable fitting)energy is proposed.The RSF model needs to repeatedly update the level set function during the iterative process,and this method of updating the fitting function shows a strong ability to segment images.But because of this,there are two problems in the RSF model:a)Repeatedly updating the fitting function requires a lot of convolution calculations,which reduces the model efficiency.b)The initialized level set function is very sensitive.Therefore,it can be considered to retain the energy term of the original RSF model and incorporate the proposed local edge fitting energy to save the ability to segment complex backgrounds and target images and improve the segmentation efficiency.The proposed local edge fitting energy is used to extract the global target boundary to attract the initial level set function to the target edge,accelerate the evolution of the level set function,and improve the model's segmentation efficiency and robustness to parameters.In essence,this is an optimized method,which is not only applicable to RSF models but also other region-based level set methods(3)An active contour model driven by local regularized fitting(LRF)energy is proposed.Existing active contour models can respond well to changes in image topology,but they are also susceptible to significant effects of gray unevenness,noise,and model parameters.In order to improve these problems,this paper proposes an active contour model driven by local regularized fitting(LRF)energy for segmenting images.This model introduces a kernel function in the proposed fitting energy,which can extract local scalable regions.Pixel information to better attract the evolution curve towards the target's boundary.At the same time,by introducing the arctangent arctan()function,the fitting energy is adjusted to a controllable range.Third,a new double-well function is proposed to avoid re-initializing the level set function and make the model more stable.Finally,a series of experiments were conducted to prove that the proposed model can not only effectively deal with the gray unevenness,but also has a higher computational efficiency,and it is also more robust to initialization.