Reference-Based Sketch Image Colorization using Augmented-Self Reference and Dense Semantic Correspondence
This work solves the problem of colorizing sketch images for applications like comics and animation, offering a user-driven approach, but it is incremental as it builds on existing reference-based methods with a novel training strategy.
The paper addresses automatic sketch image colorization using a reference image, proposing a method that uses a geometrically distorted version of the same image as a virtual reference to overcome training data scarcity. It achieves improved performance in colorization tasks, as demonstrated through quantitative and qualitative evaluations against existing methods.
This paper tackles the automatic colorization task of a sketch image given an already-colored reference image. Colorizing a sketch image is in high demand in comics, animation, and other content creation applications, but it suffers from information scarcity of a sketch image. To address this, a reference image can render the colorization process in a reliable and user-driven manner. However, it is difficult to prepare for a training data set that has a sufficient amount of semantically meaningful pairs of images as well as the ground truth for a colored image reflecting a given reference (e.g., coloring a sketch of an originally blue car given a reference green car). To tackle this challenge, we propose to utilize the identical image with geometric distortion as a virtual reference, which makes it possible to secure the ground truth for a colored output image. Furthermore, it naturally provides the ground truth for dense semantic correspondence, which we utilize in our internal attention mechanism for color transfer from reference to sketch input. We demonstrate the effectiveness of our approach in various types of sketch image colorization via quantitative as well as qualitative evaluation against existing methods.