基于高斯分布的最大似然估计框架的全局区域图像分割

Global Regional Image Segmentation Based on the Maximum Likelihood Estimation Framework of Gaussian Distribution

摘要: 利用L2范数度量模型的数据项是常用的方法之一。然而，当图像受到模糊和强度不均匀性的污染时，它的分割性能就会降低。为了解决此类问题，本文提出一种基于高斯分布的最大似然估计框架的变分水平集模型。首先，依据加性偏置场理论将观测图像进行分解得到图像的强度信息和结构信息。同时，利用高斯分布函数提取图像的纹理信息。其次，利用图像的纹理信息、结构信息及强度信息来构造演化曲线内外的拟合图像。最后，利用高斯分布的最大似然估计框架刻画背景和目标之间的差异，从而推动轮廓曲线的演变，快速找到目标边缘。实验结果表明，所提出的模型在准确性和效率方面都有显著提高，优于其他方法。定量评估结果显示，DC、JCS、P和TP的平均分割值分别是0.9845、0.9697、0.9756和0.9938。

Abstract: One of the frequently employed techniques is the use of the data term of the L2 paradigm metric model. However, its segmentation performance degrades when the image is contaminated by blurring and intensity inhomogeneity. In order to solve such problems, this paper proposes a variational level set model based on the maximum likelihood estimation framework of Gaussian distribution. First, the observed image is decomposed to obtain the intensity information and structure information of the image based on additive bias field theory. Meanwhile, the Gaussian distribution function is utilized to extract the texture information of the image. Second, the fitted image is constructed both inside and outside the evolution curve using the texture, structure, and intensity information of the image. In order to drive the evolution of the contour curve and rapidly identify the target edge, the maximum likelihood estimation framework of Gaussian distribution is finally used to illustrate the difference between the background and the target. The experimental results show that the proposed model is superior to other methods in terms of accuracy and efficiency with significant improvement. The quantitative evaluation results show that the average segmentation values of DC, JCS, P and TP are 0.9845, 0.9697, 0.9756 and 0.9938, respectively.