基于多尺度自适应加权的改进Canny算子

在Canny算法框架下,对彩色遥感图像进行多尺度滤波分析。定义多个不同尺度的高斯滤波器,在每个尺度下,分别对遥感图像RGB三个通道进行滤波和梯度计算,取三个通道的梯度最大值为该尺度下的遥感图像梯度值。根据各个尺度的滤波器对噪声的抑制能力及边缘定位能力的不同,自适应地确定相应的权值大小,然后再将这些不同尺度下检测到的梯度图像用自适应确定的权值进行加权合成最终的梯度图像。在此基础上,由非极大值抑制和双阈值处理得到图像边缘。实验结果表明,该算法比传统的Canny算子在噪声抑制和边缘定位方面具有更优的性能,适合彩色遥感图像的边缘检测。

Improved Canny Edge Detection Operator Based on Multi-scale Adaptive Weighting Algorithm

In the paper,color remote sensing image is analyzed by multi-scale Gaussian filters under Canny algorithm framework.Firstly,multiple Gaussian filters of different scale are defined,and each filter is used to filter the RGB channels of the color image,then gradient image is gained by choosing the largest gradient value as the gradient value of the specific scale.Based on the different noise restrain and edge detection ability of the filters of different scale,adaptive weights are calculated and used to form a synthesized gradient image.Finally,the edge image is obtained through non-maximum suppressing and double thresholds processing.Experiments show that the proposed algorithm renders better noise restrain and edge detection ability,which is suitable for the edge detection of color remote sensing images.