Multiscale Adaptive Edge Detector for Images Based on a Novel Standard Deviation Map

Abstract

Edge detection plays an important role in many applications, such as industrial inspection and automatic driving. However, it is difficult to effectively distinguish between faint edges and noise, which may result in losing effective edges or generating spurious edges. This will reduce the accuracy of edge detection. In addition, some parameters need to be set artificially. In the case of the fixed parameters, the overall performance of edge detection on different images is not high. The adaptivity of edge detection needs to be improved further. To solve these problems, this article proposes a multiscale adaptive edge detector for images. First, multiscale pyramid images are constructed from an input image to provide multiscale features for edge detection. At each scale, a gradient map and a novel standard deviation map are calculated based on the gradients and the statistical characteristics of the local gradient differences, respectively, to accurately distinguish the edges from the background and noise. By using these two feature maps, candidate edges are adaptively identified from the image by using pixel-by-pixel detection. Then, candidate edges at different scales are thinned and fused together based on a novel voting mechanism. Finally, a binarized edge map is obtained by using adaptive hysteresis linking. These steps make the proposed edge detector accurate and adaptive. Experiments demonstrate that the proposed edge detector achieves good performance, which is beneficial to measurement applications.