种子区域生长的高分遥感影像超像素分割

High resolution remote sensing image super-pixel segmentation based on seeded region growing

针对如何提升高分遥感影像超像素分割的精度与视觉效果问题,该文提出了一种基于种子区域生长(SRG)的超像素分割算法。该算法主要对超像素的生长过程进行了改进。在本文方法的超像素生长过程中,采用了一种充分利用光谱与形状信息的新度量标准,以有效搜索与种子适合合并的像素。该标准首先为斑块挑选在光谱上足够相似的像素,然后在这些像素中利用紧凑性异质性选择待合并的像素。为了定量评价算法的分割精度与视觉效果,定义了边界符合距离与平均斑块矩形度,并基于此发展了该文算法的参数选择策略。根据两景不同特点的高分遥感影像的超像素分割实验表明:本文方法在分割精度与视觉效果上均优于传统SRG与简单线性迭代算法。

Aiming at the problem of how to improve the accuracy and visual effect of super-pixel segmentation for high resolution remote sensing image（HRI）,a super-pixel segmentation algorithm based on seeded region growing（SRG）was proposed in this paper.This method mainly improved super-pixel growing process,in which a new metric criterion was adopted.The new metric criterion fully utilized spectral and geometric information,to effectively search for pixels that fit well into the seed pixels.This criterion first choosed the pixels with sufficient spectral similarity for the plaque,and then the compactness heterogeneity was used for further selection.In order to quantitatively evaluate segmentation accuracy and visual effect of the approach,boundary adherence distance and rectangularity of average segments were defined,and a parameter determination strategy was constructed based on these two metrics.The super-pixel segmentation experiment based on two scenes of HRI with different characteristics indicated that the proposed method outperformed the traditional SRG and the simple linear clustering approach,in terms of both segmentation accuracy and visual effect.