各向异性的多尺度边缘检测算法

Anisotropic Multi-Scale Edge Detection Algorithm

提出了一种各向异性的多尺度边缘检测算法。首先,在给定的多个尺度下分别应用各向异性的高斯滤波器获取图像的边缘强度图。然后,将所获取的多幅边缘强度图融合为一幅边缘分辨率高、拓散效应小的边缘强度图。最后,将融合的边缘强度图嵌入到Canny边缘检测算法的框架中,获得最终的边缘检测结果。创新性地提出基于“信号平均”技术的多尺度融合策略,并从理论分析和数值实验角度解释了新策略相比于现有的“几何平均”融合策略所具有的优势。实验结果表明,所提算法通过使用多尺度融合策略,有效解决了各向异性滤波器在单一尺度下存在的边缘拓散问题,能够在保持稳健噪声鲁棒性的同时,获得了比现有算法更好的边缘检测效果。

An anisotropic multi-scale edge detection algorithm is proposed. First, the edge strength maps of the input image are obtained by using a set of anisotropic Gaussian filters at given multiple scales. Then, the obtained edge strength maps are jointly used to produce one fused edge strength map that has higher edge resolution and lower edge diffusion effect. Finally, the fused edge strength map is incorporated into the framework of the Canny edge detection algorithm to generate the final result of edge detection. A new multi-scale fusion strategy based on “signal average” is proposed creatively, and the advantages of the new strategy compared with the existing “geometric average” fusion strategy are explained from the perspective of theoretical analysis and numerical experiments.Experimental results show that the proposed algorithm effectively solves the edge spreading problem of anisotropic filter in a single scale by using multi-scale fusion strategy, and can obtain better edge detection effect than the existing algorithms while maintaining robust noise robustness.