On the impact of anisotropic diffusion on edge detection

• Reviewing the main trends in Anisotropic Diffusion.
• Identifying the practical scenarios on which Anisotropic Diffusion fails.
• Quantifying the improvement of edge detection when Anisitropic Diffusion is applied.

Content-aware, edge-preserving smoothing techniques have gained visibility in recent years. However, they have had a rather limited impact on the edge detection literature compared to content-unaware (linear) techniques, often based on Gaussian filters. In this work, we focus on Anisotropic Diffusion, covering its initial definition by Perona and Malik and subsequent extensions. A visual case study is used to illustrate their features. We perform a quantitative evaluation of the performance of the Canny method for edge detection when substituting linear Gaussian smoothing filters by Anisotropic Diffusion.