Edge Detection Based on the Fusion of Multiscale Anisotropic Edge Strength Measurements

Edge detection plays an essential role in many computer vision tasks, but there is limited literature on the fusion of multiscale edge strength measurements. In this paper, we extend an edge detector using both isotropic and anisotropic Gaussian kernels in multiscale space to obtain the multiscale anisotropic edge strength measurements (AESMs). Subsequently, we propose a fusion scheme of multiscale AESMs based on geometric mean. This scheme inherits the merits of the isotropic/anisotropic Gaussian kernel based method and suppress the scale-space diffusion at the same time. Experimental results on example images in the EUSFLAT Edge Detection Competition dataset illustrate that the proposed method outperforms the widely used Canny method and the state-of-the-art isotropic/anisotropic Gaussian kernel method.