Importance-based approach for rough drawings

Heekyung Yang, Kyungha Min

Abstract: We present a framework for producing rough drawings from photographs. Depicting a scene using a series of lines is one of the most effective methods of visual communication. Our framework for rough drawing is comprised of three steps: extracting lines from images, estimating line importance, and producing strokes that express various styles. To extract lines, we employ the widely used difference-of-Gaussian filter approach to devise a fault-correcting line shift scheme. Line importance is estimated by combining gradient and saliency. To obtain an efficient saliency estimation, we propose a stochastic content-based method. Various styles of rough drawings are produced by convoluting adaptive stroke texture segments, which are prepared by sampling real stroke texture images. We test our framework on various images and compare our results with real artwork and other schemes.