Canny算子边缘检测的一种改进方法

Improved method of Canny operator edge detection

传统Canny算子在高斯滤波方差和滞后阈值的选择上需要人工干预,不具备自适应能力。高斯滤波方差的大小选择会影响到去噪和边缘保持效果,用不同尺度的形态学滤波代替高斯滤波,不仅能降低噪声影响,而且可保持边缘强度和细节;滞后阈值的选择会影响到假边缘现象的强弱和真实边缘的连续性,引入Otsu阈值法并将其推广至直方图具有多峰特点的情况,算法可根据图像自身特点选取阈值,使检测出的边缘更加连续并减少假边缘的存在。

The traditional Canny operator requires human intervention, does not have the adaptive ability themselves in the variance of the Gaussian filtering and the lag threshold.The variance of Gaussian filter affects the edge preserving and de-noising effect.Different scale morphological filter replaces the Gaussian filter, can not only reduce the noise impact,but also maintain the edge strength and details.The lag threshold affects the strength of the fake edges and the continuity of the real edges. Otsu threshold and its extension to the situation that the histogram has multiple peaks, can select the favorable threshold with the image own characteristics, so that the edge detection is more continuous and reduces the presence of fake edges.