结合图像结构信息GVF Snake模型的图像分割方法研究

发布时间：2018-10-10 14:25

【摘要】：图像分割是一种将图像中所包含的感兴趣区域与背景区域分开,并且提取出目标区域的图像处理技术。作为一种关键性的基础操作,图像分割技术已经成为图像处理领域非常重要的研究内容之一。主动轮廓(也称为Snake)模型凭借其良好的分割特性,在图像分割领域应用广泛。梯度矢量流(Gradient Vector Flow,GVF)主动轮廓模型改善了传统Snake模型对初始轮廓线位置较为敏感的问题,并且对于凹型区域的分割性能也有所提升。GVF Snake模型作为一种经典且有效的外力场改进模型也备受研究者关注。本文首先介绍了图像分割的背景和意义,并且对图像分割的概念进行了描述。然后介绍了几种经典的基础分割方法;之后重点介绍了Snake模型和GVF Snake模型的分割方法,相比于Snake模型,GVF Snake外力场的作用范围更大,分割效果更佳。然而GVF Snake模型在分割过程中也存在一些问题,针对这些不足之处,本文提出了两种改进方法。(1)GVF Snake模型在分割带有尖角的目标时,曲线很难收敛到尖角处。针对这一问题,本文提出结合角点信息GVF Snake模型图像分割方法。首先,运用基于边缘轮廓曲率的角点检测方法,检测出图像中的角点位置,并且在边缘线上和角点处对GVF场进行局部修正,然后结合角点信息给出局部角点力,最后将角点力与修正后的GVF场相结合得出一种新的外力场。实验证明,本文改进的GVF Snake模型能够更好的收敛到图像的尖角处。(2)GVF Snake模型相比于传统的Snake模型在分割凹型边界性能方面有了一定的提升。然而,对于深凹区域的分割,GVF Snake模型仍然很难收敛到深凹区域底部,并且GVF Snake模型对于噪声的鲁棒性以及边缘保护方面也存在不足。针对这些问题,根据广义GVF(Generized GVF,简称为GGVF)Snake模型,本文提出了基于图像结构信息各项异性GGVF(Image Structure Anisotropic GGVF,简称ISAGGVF)Snake模型。首先,求出图像的结构张量,然后根据图像结构张量构建各项异性扩散矩阵。最后将GGVF Snake模型中的各项同性扩散替换成各项异性扩散矩阵。这样外力场中的扩散项就是根据图像的结构信息自适应调节扩散系数。实验证明,本文改进的模型能够准确收敛到深凹底部,并且对于噪声具有一定的鲁棒性。
[Abstract]:Image segmentation is an image processing technique that separates the region of interest from the background region and extracts the target region. As a key basic operation, image segmentation technology has become one of the most important research contents in the field of image processing. Active contour (also known as Snake) model is widely used in image segmentation field because of its good segmentation characteristics. Gradient vector flow (Gradient Vector Flow,GVF) active contour model improves the sensitivity of the traditional Snake model to the position of the initial contour. And the segmentation performance of concave region is also improved. As a classical and effective improved model of external force field,. GVF Snake model has attracted much attention. This paper first introduces the background and significance of image segmentation, and describes the concept of image segmentation. Then several classical basic segmentation methods are introduced, and the segmentation methods of Snake model and GVF Snake model are emphasized. Compared with the Snake model, the external force field of, GVF Snake is larger and the segmentation effect is better than that of Snake model. However, there are some problems in the segmentation of GVF Snake model. In view of these shortcomings, two improved methods are proposed in this paper. (1) the curve of GVF Snake model is difficult to converge to the sharp angle when it is used to segment the target with sharp angle. In order to solve this problem, a method of image segmentation based on corner information GVF Snake model is proposed in this paper. Firstly, the corner position in the image is detected by using corner detection method based on the curvature of the edge contour, and the GVF field is locally corrected on the edge line and corner, and then the local corner force is given by combining the corner information. Finally, a new external force field is obtained by combining the corner force with the modified GVF field. Experimental results show that the improved GVF Snake model can converge better to the sharp corner of the image. (2) compared with the traditional Snake model, the performance of the) GVF Snake model is improved in the concave boundary segmentation. However, the segmented, GVF Snake model for deep concave region is still difficult to converge to the bottom of deep concave region, and the robustness of GVF Snake model to noise and edge protection are also insufficient. In order to solve these problems, according to the generalized GVF (Generized GVF, referred to as the GGVF) Snake model, this paper presents the heterosexual GGVF (Image Structure Anisotropic GGVF, ISAGGVF) Snake model based on the image structure information. Firstly, the structure Zhang Liang of the image is obtained, and then the heterosexual diffusion matrix is constructed according to the image structure Zhang Liang. Finally, the homogeneity diffusion in the GGVF Snake model is replaced by the heterosexual diffusion matrix. Thus the diffusion term in the external force field adaptively adjusts the diffusion coefficient according to the structure information of the image. Experimental results show that the improved model can converge to the deep concave bottom accurately and is robust to noise.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】