基于Delaunay划分结合EM\MPM算法的图像分割方法

Combining the EM\MPM and Delaunay Tessellation for Image Segmentation

为了克服传统基于区域的图像分割方法对图像初始划分完全随机进而导致算法效率低下的缺点,本文提出了一种基于Delaunay划分并结合最大期望值(Expectation Maximization,EM)和最大边缘概率(Maximization of the Posterior Marginal,MPM)算法的图像分割方法。该方法首先提取图像特征点,并把特征点集作为构建Delaunay三角网的基础点集。利用Delaunay三角网的构建将影像划分成众多彼此连接的超像素,并假设这些超像素内的像素灰度值服从同一独立的正态分布,基于此完成特征场模型的建立,再运用EM\MPM方法分别模拟特征场模型和分割影像。为了验证本文提出的算法能够有效地分割图像,分别对模拟图像和真实图像进行分割测试,并和经典的初始划分完全随机的超像素影像分割算法进行对比,测试结果定性和定量地表明了该方法的有效性和准确性。

In order to overcome the shortcomings that the completely random initial partition leads to the inefficiency of the traditionalimage segmentation methods based on region of image, a method based on Delaunay partition combined with Expectation Maximizationand Maximization of the Posterior Marginal algorithm is proposed. Firstly, the feature points of the image are extracted, and the featurepoints set is used as the fundamental points set of the Delaunay triangulation to construct the Delaunay triangulation. Delaunay triangulation divides the image into several super pixels which are connected each other, assuming that the gray values of pixels in these regions obey the same independent normal distribution, so as to construct the characteristic field model and then combine with the EM/MPM algorithm for image segmentation and model parameter estimation. To verify the proposed algorithm is able to segment images effectively, the simulated image and the real images are segmented and compared with the classical image segmentation algorithm basedsuper pixels which have the completely random initial partition. The qualitative and quantitative results show the validity and accuracyof the proposed method.