Deep learning for anime style transfer

Der Lor Way, Wei Cheng Chang, Zen Chung Shih

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Some artificial systems based on a deep neural network create artistic images of high perceptual quality. However, it is usually suitable for use in abstract styles. The performances of existing style transfer algorithms on anime style are not very satisfactory, because it is either not sufficiently stylized or distorted severely in comic characters' domain. In this paper, we propose a novel anime style transfer algorithm using deep neural network, which treats foreground and background differently. Moreover, our method also could transfer the style for video with a style image. We combine semantic segmentation and spatial control to transfer the specified style to the specified area. By designing the initial image and the loss function. Users could adjust the feature weights of different regions to maintain the artistic conception of the target style, and combine optical flow to ensure frame coherence in a video. Finally, some experimental results demonstrate the effectiveness of our proposed method.

Original languageEnglish
Title of host publicationICAIP 2019 - 2019 3rd International Conference on Advances in Image Processing
PublisherAssociation for Computing Machinery
Pages139-143
Number of pages5
ISBN (Electronic)9781450376754
DOIs
StatePublished - 3 Nov 2019
Event3rd International Conference on Advances in Image Processing, ICAIP 2019 - Chengdu, China
Duration: 8 Nov 201910 Nov 2019

Publication series

NameACM International Conference Proceeding Series

Conference

Conference3rd International Conference on Advances in Image Processing, ICAIP 2019
CountryChina
CityChengdu
Period8/11/1910/11/19

Keywords

  • Anime Style
  • Deep Learning
  • Motion Estimation
  • Semantic Segmentation
  • Style transfer

Fingerprint Dive into the research topics of 'Deep learning for anime style transfer'. Together they form a unique fingerprint.

Cite this