图像多尺度边缘检测及图像多尺度分割研究

Multi-scale Edge Detection and Multi-scale Segmentation of Imagery

提出了一种改变Canny阈值的图像多尺度边缘检测方法,该方法有效保持了边缘位置精度,不同尺度下的边缘之间不存在位移;利用多尺度边缘特征与多尺度光谱特征,实现了融合两种特征的图像多尺度分割,能够有效降低过分割和欠分割现象,使得分割结果更忠实于原地物。

The scale is a basic issue of geography field.Many algorithms of multi-scale image segmentation,based on spectral features are put forward,but the multi-scale image segmentation,fusing multi-scale edge features and multi-scale spectral features are rarely researched.This paper presents a method to extract multi-scale edges with changing the Canny thresholds,which is effective to preserve the accuracy of edge location,and the edges between different scales don′t shifted.This multi-scale edge features and multi-scale spectral features are integrated to finish the multi-scale image segmentation.The results show that our methods can effectively reduce over-segmentation and under-segmentation,and the segmentation results are more faithful to the original objects.