Font Size: a A A

Reseach On Optimization And Level Set Methods For Image Segmentation

Posted on:2015-12-27Degree:DoctorType:Dissertation
Country:ChinaCandidate:H WangFull Text:PDF
GTID:1108330473456016Subject:Applied Mathematics
Abstract/Summary:PDF Full Text Request
Image segmentation is a hot topic in image processing and computer vision, which is always concerned by scholars. In the field of practical production and living, image segmentation can be applied to many domains such as object identification, image analysis,medical imaging and remote sensing image. Based on the characteristic, the realization of image segmentation is to divide a given image as different regions and further extracts the interested object from the background. However, the solution of image segmentation is not unique, which is an ill-posed problem. There exist many methods for image segmentation, but so far there is not a common and efficient approach, and even not a clear criterion for classifying these methods. In the current image segmentation methods, there still exist that the segmentation accuracy is not desirable, and the efficiency is low, and the commonality is bad, and so on. With the development of changing with each passing day for the electronic computer, the field of image segmentation meets the updating problems and requirements, and faces great challenges as well. Thus, it needs that we continue to pursue and explore new segmentation approaches. Therefore, the research on image segmentation is significant. Due to the good performance, the image segmentation approaches based on the optimization and level set methods have become an increasingly prominent attention of many scholars. This thesis aims at improving the speed and accuracy of image segmentation, and has researched the optimization and level set methods of image segmentation. The main contents are as follows:1. Based on typical active contour models, such as the GAC model and the CV model, a fusing level set active contour model is proposed. This model effectively includes the boundary and region information and the numerical implementation of the level set evolution equation is simple. The model analysis and numerical experiments demonstrate the effectiveness of this model.2. By employing the statistical intensity information, an active contour model is proposed, which is based on the local and global Gaussian distribution fitting energies.Besides, a corresponding algorithm is designed. This model includes the local and global region information of the image, which is superior to the existing region-based active contour models. Specially, we design a new algorithm. Based on the change of the active contour location, this algorithm makes that the weight of the local and global region fitting energies of the model in the total energy functional is adaptively changed, which is benefit to speeding up the convergence rate and improving the segmentation accuracy.Numerical experiments verify that this algorithm can obtain satisfactory segmentation results.3. Because the C-V model can not effectively deal with the image with intensity inhomogeneity, and during the segmentation process, it needs to re-initialize the level set function, we introduce the local region information and propose a local region C-V model. Further, by employing the Laplacian operator, we design an efficient operator split method. Meanwhile, we extend this method to vector-valued image segmentation.Algorithm analysis and numerical experiments show the effectiveness of this method.4. Aiming at a kind of the image with thick boundary, an image segmentation model is presented, which can be used to extract the object and background, respectively. Without considering the boundary, this model uses two curves, and specifically one is used to extract the object, and the other is used to segment the background. In addition, this model also includes the distance restraint term and the level set regularized term. Numerical experiments demonstrate that this model can obtain desirable segmentation results.5. Adaptive weighting parameter selections are researched. Aiming at the difficulty of choosing a fit weighting parameter in the existing model based on the local and global region information, an adaptive weighting parameter selection method is proposed. With the evolution of the level set, this method efficiently realizes the transformation of the dominant energy functional from the global region information to the local region information, which is benefit to initializing the active contour and improving the image segmentation accuracy. Numerical experiments demonstrate that this method can obtain desirable segmentation results. On the other hand, based on the typical C-V model, the boundary information is introduced, which realizes the integration of the global region information and the boundary information, and an adaptive weighting parameter method is proposed. This method effectively balances the effect of the boundary information and the global region information during the energy minimization for image segmentation. Numerical experiments show that this method can adaptively choose the weighting parameter between the boundary information and the region information, which is fit to obtain accurate segmentation results.6. By introducing the statistical information of the image, and aiming at the nonconvex property of the local and global region active contour model, we propose a global minimization hybrid active contour model. This model efficiently solves the problem of the non-convex property of the original model, and relying on the optimization method,this model can converge to a global minimum. In numerical implementation, a fast numerical approach is introduced, which is faster than the conventional gradient descent flow method. Numerical experiments demonstrate that this model can obtain accurate segmentation results and especially this model can be used to segment oil spill images.7. Based on the global and local region active contour model, an efficient two-stage segmentation method is proposed. This method is less sensitive to the initialization and improves the segmentation accuracy. Numerical experiments show that this method can obtain desirable segmentation effectiveness.
Keywords/Search Tags:image processing, image segmentation, optimization, level set, active contour
PDF Full Text Request
Related items