水平集方法在图像分割中的应用

发布时间：2018-06-22 20:33

本文选题：图像分割 + 水平集方法　；参考：《中央民族大学》2017年硕士论文

【摘要】：随着计算机科学技术的发展,数字图像被广泛应用于各个领域,利用计算机进行图像分割是将图像中的有用信息提取出来,从而对相关信息进行分析。水平集方法在图像分割和计算机视觉领域有很广泛的应用,具有非常好的分割性能。本文对常用的水平集分割模型:蛇模型、GAC模型、M-S模型、C-V模型、LBF模型的基本分割原理、模型的构建、模型的优缺点进行了全面的分析和介绍,并针对其存在的问题进行了改进。在传统的水平集方法中,水平集函数需要保持符号距离函数,而现有的模型均需要对水平集函数进行重新初始化,使其保持符号距离函数,这样会引起数值计算的错误,最终破坏演化的稳定性。另外,部分模型只适用于灰度值较为均匀的图像,对灰度值不均匀的图像不能进行理想的分割。针对这些问题,本文结合C-V模型和LBF模型的思想,提出了两种新的图像分割模型。1.新型的四相水平集图像分割模型,该模型结合了 C-V模型的思想,应用两个水平集函数对灰度不均匀的图像进行分割,特别是其正则项被定义为一个双势函数,具有向前向后扩散的作用,使水平集函数在演化过程中保持为符号距离函数,避免了重新初始化的过程。最后对该模型进行数值仿真,对简单的cube图,Brain图,子宫肌瘤CT图等进行分割,实验表明了新模型能够更好的分割灰度不均匀的图像,轮廓清晰,计算速度较快,充分证明了新模型的可操作性和有效性。2.新型的LBF模型,该模型结合了 LBF模型的思想和C-V模型的思想,利用局部和全局的信息对图像进行分割,该模型的正则项仍然定义为一个双势函数,确保水平集函数在演化过程中保持为符号距离函数,应用变分水平集方法,通过双势函数得到最小化能量泛函的梯度下降流方程。最后对其进行数值仿真,实验发现,新模型能够对复杂的灰度不均匀的医学图像进行很好的分割,且利用双势函数的优点,避免了水平集函数的重新初始化过程,使其时刻保持为符号距离函数,充分证明了新模型的可行性。
[Abstract]:With the development of computer science and technology, digital image is widely used in various fields. Image segmentation by computer is to extract the useful information from the image and analyze the relevant information. Level set method is widely used in image segmentation and computer vision. In this paper, the basic segmentation principle, the construction of the model, the advantages and disadvantages of the LBF model are analyzed and introduced in detail, and the existing problems are improved according to the common horizontal set segmentation model: the serpent GAC model and the M-S model and the C-V model and the LBF model. In the traditional level set method, the level set function needs to maintain the symbolic distance function, but the existing models need to reinitialize the level set function to keep the symbolic distance function, which will lead to the error of numerical calculation. Finally destroy the stability of evolution. In addition, part of the model is only suitable for the image with more uniform gray value, and the image with uneven gray value can not be segmented perfectly. In order to solve these problems, two new image segmentation models, I. e., C-V model and LBF model, are proposed in this paper. A new four-phase horizontal set image segmentation model, which combines the idea of C-V model, uses two level set functions to segment an image with uneven grayscale, especially when its regular term is defined as a double potential function. It has the function of forward and backward diffusion, which keeps the level set function as the symbolic distance function in the evolution process, thus avoiding the process of reinitialization. Finally, the numerical simulation of the model is carried out, and the simple cube diagram brain map, hysteromyoma CT image and so on are segmented. The experiment shows that the new model can segment the uneven gray image better, the contour is clear, and the calculation speed is faster. The maneuverability and validity of the new model are fully proved. A new LBF model, which combines the idea of LBF model with the idea of C-V model, uses local and global information to segment the image. The regular term of the model is still defined as a double potential function. It is ensured that the level set function remains a signed distance function in the evolution process. By using the variational level set method, the gradient descent flow equation for minimizing the energy functional is obtained by using the double potential function. Finally, the numerical simulation results show that the new model can segment the medical image with complex grayscale heterogeneity and avoid the reinitialization of the level set function by using the advantage of the double potential function. The new model is proved to be feasible by keeping it as a symbolic distance function.
【学位授予单位】：中央民族大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】