Edge detection is the front-end processing stage in most computer vision and image understanding systems. Among various edge detection techniques, Canny edge detector is the one of most commonly used. In this paper a modified Canny edge detection technique focusing on change of the Sobel operator is proposed. Instead of convolution kernels, the weighted least squares method is utilized to calculate the horizontal and vertical gradient. Experimental results show that the new detector can detect some edges which are not observed in the results using the Canny edge detector.

中文翻译：

---

基于加权最小二乘的改进型Canny边缘检测器

边缘检测是大多数计算机视觉和图像理解系统中的前端处理阶段。在各种边缘检测技术中，Canny边缘检测器是最常用的一种。本文提出了一种改进的Canny边缘检测技术，其重点是Sobel算子的变化。代替卷积核，加权最小二乘法被用来计算水平和垂直梯度。实验结果表明，使用Canny边缘检测器，新检测器可以检测出一些未在结果中观察到的边缘。