Abstract: The edge detection algorithm IAGK combining isotropic and anisotropic Gaussian filters has edge stretching effect, which causes false edges at complex edges. The introduction of isotropic Gaussian derivative filters also declines the robustness of the algorithm. At the same time, the anisotropic factor of IAGK algorithm is non-optimal value, and theoretically does not have optimal SNR and positioning performance. In this paper, the optimal anisotropic factor is selected based on the automatic anisotropic Gaussian kernel, and the edge strength correction factor is used to modify the edge strength formula of the IAGK algorithm to suppress the generation of false edges and the influence of noise. The proposed edge detection algorithm is tested on classical edge detection datasets. The experimental results show that, compared with Canny, AAGK and IAGK algorithms, the proposed algorithm has better noise robustness and better weak edge detection capability. The pseudo edge effect can also be further reduced. In the case of noise, the figure of merit (FOM) of the proposed algorithm is 3%, 4% and 7% higher than the Canny, AAGK and IAGK algorithms, respectively.

Key words: edge detection, anisotropy factor, edge stretching, ROC curve, figure of merit(FOM), edge strength map