Abstract: With the development of remote sensing technology, high-resolution remote sensing imagery’s content becomes more and more complicated. Information presentation is more and more precise. If these image segmentation or classification methods suitable for medium-sized and low resolution imagery are applied to high-resolution imagery directly, result is always unsatisfactory. On the basis of the algorithm of the traditional region merging, high-resolution imagery is carried out quadruplex optimization, adopted four neighborhood region tactics to plot out different pixel pairs, which are merged into bigger region under different scales and strictly establishing merging criterion. The area threshold is determined, and region smaller than threshold will be merged. Segmentation imagery is obtained. Experiment proves that the algorithm realizes extraction of object to some extent, reducing calculation work of traditional regional merging algorithm, taking the spatial characteristic of pixel into consideration. Meanwhile, it possesses good  noise-resistance ability, meeting the demands of dealing with high-resolution remote sensing image with intense noise.

Key words: high-resolution remote sensing imagery, multi-scale segmentation, statistical region merging, four-neighborhood regional strategy, polygonal object

摘要： 随着遥感技术的发展，高分辨率遥感图像内容变得越来越复杂，信息表现更加精细，将某些中、低分辨率图像的图像分割、 分类算法直接应用到这类图像上时，其效果往往不理想。在传统区域增长算法的基础上，进行了四重优化，采用四邻域策略组合成像素对，严格化合并准则，在不同尺度下将像素对进行合并，得到较好的分割影像。实验证明，该算法较大程度上实现了目标的提取，减少了传统区域合并算法的计算量，与原算法相比较更多地考虑了像素的空间特性，具有很好的抗噪能力，能够适应于大数据量高分辨率遥感影像的多尺度分割。

关键词: 高分辨率遥感影像, 多尺度分割, 统计区域增长, 四邻域策略, 多边形对象