中文摘要：

为了解决高分辨率遥感影像中相同地物目标异质性和空间破碎性增大及不同地物目标的相似性增强所带来的分割新问题,该文提出一种融入空间关系的高斯混合模型（GMM）高分辨遥感影像监督分割方法。该方法首先按分割区域进行监督采样,并通过最小二乘法进行直方图拟合,对影像中的每个类别区域建立GMM用来精确表征高分辨遥感影像每个分割区域复杂的地物光谱特征;然后在GMM的概率测度域融入空间关系,使每个像素的区域所属由该像素邻域窗口内所有像素概率测度共同决定,以刻画高分辨率遥感影像中像素间的空间相关性;最后按照最大概率测度原则完成对高分辨率遥感影像的分割。为了验证文中算法的可行性与有效性分别对合成影像及真实高分辨率遥感影像进行分割实验,并和经典的FCM方法及HMRF-FCM方法进行对比,定量与定性的结果证明了文中方法能够提高分割精度。

英文摘要：

This paper proposes a supervised image segmentation algorithm for high resolution remote sensing images by introducing the Gaussian Mixture Model （GMM） with spatial relationship in order to solve the problem of the increasing dissimilarity in the same object and the decreasing of dissimilarity between two different objects. The proposed algorithm takes samples according to the segmentation areas and uses the least squared method to fit the histogram. GMMs are established to describe the complex spectral characteristic in each area accurately. Then spatial relationships are taken consider into the probability measures in GMMs to make the dissimilarities of pixels in a window is determined by all the pixels in the same window. Overall the GMMs can describe the spatial relationships between the pixels in high resolution remote sensing images. Finally the segmentation result is obtained by maximum probability principle. To verify the feasibility and the effectively of the proposed algorithm, the algorithm is performed on real high resolution remote sensing and synthetic images and compared the results with that of FCM and HMRF-FCM based segmentation algorithm. Qualitative and quantitative results prove that the proposed algorithm could improve the accuracy of segmentation.