A Stylization Method for Edge Map Extraction

This paper presents a non-photorealistic rendering technique that automatically extracts stylized edge map from 2D images, aiming to facilitate effective visual communication. We introduce an anisotropic DoG (difference-of-Gaussians) filter that can extract a set of coherent, smooth, and stylized lines from 2D images. These lines effectively capture and convey important shapes and details in the image. In contrast to conventional edge detector canny method, the edge map generated by proposed algorithm has more hand-painting characteristic. Experimental results demonstrate that this technique produces stylized lines at the same time preserves significant geometric features.