基于多尺度的区域生长的图像分割算法

Medical image segmentation algorithm based on multi-scale region growing

针对医学图像的特点,提出了一种多尺度区域生长分割算法。该算法首先利用高斯滤波器对原图像进行滤波处理,之后利用区域生长算法分别对原图像与平滑图像进行分割操作,最后将两个分割图谱进行比对,获得最终的分割结果。进行区域生长算法时,从背景区域选择一个像素作为初始种子点进行区域生长。该方法的优势为噪音具有较好鲁棒性,初始种子点选取规则简单。该方法同样适合于其他背景简单、目标区域复杂的图像分割情形。为了选取合适分割阈值,提出了最大梯度概念。利用图像的最大梯度矩阵的统计特性,将阈值选取问题转化为求最小值问题。实验表明,该方法能够准确地获得医学图像的分割结果。

Considering the characteristics of medical images,a multi-scale region growing segmentation algorithm is proposed.First,the original image is smoothed by a Gaussian filter.Then,the region growing segmentation is performed on both the original image and the smoothed image.Finally,the two obtained segmentation maps are compared to get the final segmentation.In the process of region growing segmentation,apixel in the background region is selected as the initial seed for region growing.The advantages of the proposed algorithm are the robustness to noises and easiness of initial seed selection.This method is also suitable to other segmentation applications,in which background region is simple but target region is complex.In order to select an appropriate threshold,the concept of Maximum Gradient Transform(MGT)is proposed.The issue of threshold selection is converted into a minimization problem with the assistance of the statistical properties of the transformation matrix.Experiment results show that the proposed algorithm can obtain accurate medical image segmentation results.