Research On Active-Contour-Model-based Image Segmentation Algorithm

Image segmentation plays an important role in the areas of image processing and computer vision.Among them,the segmentation method based on active contour has developed greatly in recent years because of its effective treatment on complex images.Now it has achieved significant effect in the fields of medicine,military,industry,etc.This article conducts a study on the segmentation methods and focuses on discussing the active-contour-based method.Firstly,the segmentation methods are divided into edge-based and region-based,whose characteristics are summarized respectively.Then,the active-contour-based methods are studied and several models are compared.At last,the Local Binary Fitting model and the Local Gaussian Distribution Fitting model are improved separately.The specific work is described as follows.1.The LBF model just uses the local image information and tends to fall into a local extreme value,which usually leads to the failure of evolution,thus an improved algorithm is proposed.This method introduces the Distance Regularized Level Set Evolution model based on edge information,which makes full use of the advantages of both models,and converges the contour curve to the position near the target quickly to implement fine segmentation.The simulation result shows that the method can improve the speed of the original model with guarantee of accuracy and reduces the sensitivity to the initial contour at the same time.It promotes the performance compared with the primary two models.2.Because the LGDF model needs a manual initial contour and it is sensitive to the position of the initial contour,an improved method is proposed by using the clustering algorithm and an improved kernel function.By applying this method,the ideal initial contour can be obtained in the unsupervised-learning condition,which can be substituted into the level set frame of the model with an improved kernel for iterative evolution.Besides,because the LGDF model needs to end the evolution manually,and the setting of the suitable parameters need lot of experiments,an improved method is proposed to end the evolution automatically.It combines the intermediate results of the iterative evolution.When the contour evolves to the specified target boundary,the segmentation is stopped automatically at the appropriate time.Simulation results show that the proposed method can promote the contour curve's convergence speed,and avoid the possibilities of deviation and failure caused by the unreasonable manuallabelled contour and the invariable local minimum.