一种基于零交叉的图像边缘检测的改进方法

理论上，边缘像素应该是二阶梯度图像产生零交叉时的位置对应的像素，但对于斜坡边缘这一类灰度变化非常缓慢的情形，单纯的零交叉方法并不能得到合适的边缘。为解决这类问题，在高斯拉普拉斯边缘检测算子的基础上，结合了梯度图像二值化，边界跟踪和细化技术，从而得到了具有较高定位精度的单像素宽的边缘图像。

An Improved Approach of Image Edge Detection Based on Zero-crossing

In theory, edge pixels should be location where there exists zero-crossing in second-order grad image. But under the situation of sloping edge, gray level changes very slowly, one can not get effective edge by the zero-crossing method. For solving this issue, the paper combines the techniques of grad image binaryzation, contour tracing and edge thinning on the basis of Gauss-Laplace edge operator, and obtaines one-pixel-wide edge image with high location precision.