Sketch Image Style Transfer Based on Sketch Density Controlling.

Sketch Image Style Transfer aims to stylize the sketch images from given art images to make them look like the same artistic styles while still persevering the original sketch contents. Previous methods generally disentangle the content and style of reference image and transfer the style to sketch image. However, the textures or the painting strokes of the reference art image could be a part of content as well as style. It is difficult to decide how much it should be involved for sketch image style transferring. In this paper, we propose a novel sketch image style transfer method to use the sketch density to control it, reaching the common sense that simpler sketch images own richer textures of reference style images after stylization and otherwise the opposite. The proposed model is built upon the general content and style encoding and fused decoding architecture, but adds sketch density extraction to obtain the density level of input sketch image, and uses it to control texture information during transferring. With these special designing, our model is more flexible for inputting multi-density sketch images and reasonable for deciding the degree to transfer the texture styles of reference image. Experimental results on different datasets demonstrate the effectiveness and superiority of our method for sketch image style transfer.