论文部分内容阅读
针对传统方法对过分割未作处理以及边界语义上下文信息不强的缺点,对水陆场景图像进行区域分类,得到描述大尺度地物分布的分类结果.该方法分为3个过程:首先对图像进行纹理特征聚类,得到初始分割结果;然后对分割图像进行区域合并;最后利用边缘密度作为区域的特征,通过对大量样本的统计分析,确定一个经验值作为分类阈值,根据阈值将各个区域标记为水域或陆地,从而得到原始图像的水陆分布.与传统方法的对比实验表明:本方法保证语义正确的同时,计算效率有较大提高,实现了场景的快速分类.
In order to overcome the shortcomings of over-segmentation and low-level semantic contextual information in traditional methods, the classification of land-surface images is carried out, and the classification results describing the distribution of large-scale features are obtained.This method is divided into three steps: firstly, Finally, the edge density is used as the feature of the region. Through statistical analysis of a large number of samples, an empirical value is determined as the classification threshold, and each region is marked as Water or land, so as to obtain the original image of the land-water distribution.Compared with the traditional methods of experiments show that: the method to ensure the correct semantic, while the computational efficiency has greatly improved, to achieve the rapid classification of the scene.