Abstract: Because traditional square-shaped edge detectors based on gradient only contain few directions （generally 2 or 4）, edges are not detected from multi-resolution point of view. Therefore so much edge information from other directions is lost. Aiming at the above problem, an edge descriptor, called Varied Local Edge Pattern （VLEP）, with multi-scale and multi-resolution properties is proposed, and is applied to edge detection. Main idea about this method involves that image is firstly smoothed by Gaussian filter. Then one or more groups of VLEP descriptors are convolved with the processed image, and edge intensity and edge gradient value are obtained. Finally edges are extracted by adaptable gradient thresholds and binaryzation. In addition, weighted fusion idea is considered when multiple groups of VLEP descriptors are used, so that richer edge information is obtained. Experimental results show that the proposed edge detection method achieves better performance than other state-of-the-art edge detection methods.

Key words: edge detection, multi-scale, multi-resolution, Varied Local Edge Pattern （VLEP）, weighted fusion

摘要： 由于传统基于梯度的方形边缘检测算子包含边缘方向过少（一般为2个或4个方向），因此无法从多分辨率角度检测边缘，进而会丢失其他方向的边缘信息。针对上述问题，提出一种具有多尺度、多分辨率特性的边缘检测算子，称为可变局部边缘模式（Varied Local Edge Pattern，VLEP）算子，并用来提取图像边缘信息。算法主要思路包括，将图像经过高斯滤波器平滑，使用一组或多组VLEP算子与滤波后的图像进行卷积，得到边缘强度，从而获得边缘梯度值，最后设置适当的梯度阈值，对梯度图像进行二值化处理，完成图像的边缘检测。此外，当多组VLEP算子被同时使用时，考虑结合加权融合思想，以便获得更加丰富的边缘信息。实验结果表明，提出的边缘检测算法比其他经典的方法具有更好的边缘检测效果。

关键词: 边缘检测, 多尺度, 多分辨率, 可变局部边缘模式, 加权融合