基于水平集的灰度不均匀图像分割算法研究

发布时间：2018-01-07 15:31

本文关键词：基于水平集的灰度不均匀图像分割算法研究　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：图像分割技术是图像处理领域中的一项重要研究内容,也是图像分析与目标识别的重要步骤。迄今为止,国内外学者提出了多种图像分割方法,然而由于图像的复杂性和多样性,图像分割依然是一项重要而富有挑战性的研究课题。在所有的图像分割算法中,基于曲线演化理论的水平集算法是受到了极大的关注。其利用了轮廓曲线动态演化的思想,并且具有严谨的数学理论基础,能够解决很多其他分割方法难以解决的问题。本论文深入研究了一些经典的水平集模型,针对存在的缺陷和不足提出了一些改进,并最终能够获得理想的分割效果。具体的研究工作如下:(1)针对边缘型水平集模型对曲线初始轮廓比较敏感的问题,给出了一种基于空间模糊聚类的边缘型水平集分割模型。首先采用空间模糊聚类算法对图像预分割,然后根据预分割的结果对边缘型水平集演化模型中水平集函数进行初始化,并加入使用双阱势函数的距离规则项来避免在演化过程中水平集函数周期性初始化的问题。该算法引入了图像空间域信息,克服了初始轮廓与参数均需要手动设定的缺点,并由于确定的初始位置,有效缓解了边缘型模型对初始轮廓敏感的问题,使得分割结果更加准确。通过边界十分模糊的乳腺肿块图像对该算法进行验证。经过试验验证该算法能够自动初始化并正确分割图像。(2)针对LBF模型对初始轮廓比较敏感且容易陷入局部最优的问题,给出了一种引入全局信息的局部区域型水平集分割模型。该模型将提供全局信息的C-V模型和提供局部信息的LBF模型通过局部熵结合起来,构建能量泛函,同时给出水平集演化的理论推导和数值求解。有效解决了 LBF模型对轮廓初始化敏感且容易陷入局部最优的问题,同时也可以解决C-V模型不能处理灰度不均匀图像的问题,并且可自动设置权重。最后通过灰度不均匀图像验证该算法的有效性。(3)针对LIC模型对图像修正的偏置场没有实质性约束(偏置场平滑且缓慢变化),导致偏置场修正结果以及图像分割结果不是十分理想的问题,给出了一种基于乘法优化的局部聚类水平集图像分割模型。通过一组平滑的线性基函数对偏置场进行拟合,以在理论上保证偏置场的光滑性。将图像分割与偏置场修正融合在一个能量泛函框架中,并给出水平集演化方程的理论推导和数值求解。该算法有效提高了 LIC模型的分割精度,并对偏置场进行了约束。最后通过合成图像与医学灰度不均匀图像进行试验,验证了该算法的有效性。
[Abstract]:Image segmentation is an important research content in the field of image processing, and it is also an important step in image analysis and target recognition. So far, many kinds of image segmentation methods have been proposed by domestic and foreign scholars. However, due to the complexity and diversity of images, image segmentation is still an important and challenging research topic. The level set algorithm based on curve evolution theory has attracted much attention. It makes use of the idea of dynamic evolution of contour curve and has a rigorous mathematical theory foundation. Can solve many other segmentation methods difficult to solve. This paper in-depth study of some classical level set model, aiming at the shortcomings and shortcomings of some improvements. Finally, the ideal segmentation effect can be obtained. The specific research work is as follows: 1) aiming at the problem that the edge level set model is more sensitive to the initial contour of the curve. An edge level set segmentation model based on spatial fuzzy clustering is presented. Firstly, the spatial fuzzy clustering algorithm is used to presegment the image. Then the level set function in the evolution model of edge level set is initialized according to the result of pre-segmentation. The distance rule term of the double well potential function is added to avoid the problem of periodic initialization of the level set function in the evolution process. The algorithm introduces the spatial domain information of the image. It overcomes the shortcoming that the initial contour and parameters need to be set manually, and because the initial position is determined, the problem that the edge model is sensitive to the initial contour is effectively alleviated. The segmentation result is more accurate. The algorithm is verified by the breast mass image with very fuzzy boundary. The experiment shows that the algorithm can initialize automatically and segment the image correctly. In order to solve the problem that LBF model is sensitive to initial contour and is prone to fall into local optimum. In this paper, a local region-type horizontal set segmentation model with global information is presented, which combines the C-V model with the local information model and the LBF model with local information through local entropy. The energy functional is constructed, and the theoretical derivation and numerical solution of the level set evolution are given, which effectively solves the problem that the LBF model is sensitive to contour initialization and is prone to fall into local optimum. At the same time, it can also solve the problem that C-V model can not deal with uneven grayscale images. And the weight can be set automatically. Finally, the validity of the algorithm is verified by grayscale non-uniform image. (3) there is no material constraint (smooth and slow change of bias field) for the offset field modified by LIC model. The results of bias field correction and image segmentation are not very ideal. A local clustering level set image segmentation model based on multiplication optimization is presented. A set of smooth linear basis functions is used to fit the bias field. In order to ensure the smoothness of bias field theoretically, image segmentation and offset field correction are fused into an energy functional framework. The theoretical derivation and numerical solution of the evolution equation of the level set are given. The algorithm improves the segmentation accuracy of the LIC model effectively. The bias field is constrained. Finally, the validity of the proposed algorithm is verified by the experiments of synthetic images and medical grayscale non-uniform images.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】