Vectorization of Raster Manga by Deep Reinforcement Learning

  • 2021-10-10 15:52:38
  • Hao Su, Jianwei Niu, Xuefeng Liu, Jiahe Cui, Ji Wan
  • 58

Abstract

Manga is a popular Japanese-style comic form that consists of black-and-whitestroke lines. Compared with images of real-world scenarios, the simplertextures and fewer color gradients of mangas are the extra natures that can bevectorized. In this paper, we propose Mang2Vec, the first approach forvectorizing raster mangas using Deep Reinforcement Learning (DRL). Unlikeexisting learning-based works of image vectorization, we present a new viewthat considers an entire manga as a collection of basic primitives "strokeline", and the sequence of strokes lines can be deep decomposed for furthervectorization. We train a designed DRL agent to produce the most suitablesequence of stroke lines, which is constrained to follow the visual feature ofthe target manga. Next, the control parameters of strokes are collected totranslated to vector format. To improve our performances on visual quality andstorage size, we further propose an SA reward to generate accurate stokes, anda pruning mechanism to avoid producing error and redundant strokes.Quantitative and qualitative experiments demonstrate that our Mang2Vec canproduce impressive results and reaches the state-of-the-art level.

 

Quick Read (beta)

loading the full paper ...