Semantically meaningful vectorization of line art in drawn animation

Information

• Publication Type: Master Thesis
• Workgroup(s)/Project(s):
  • Rendering and Modeling
• Date: March 2024
• Date (Start): May 2022
• Date (End): 11. March 2024
• Second Supervisor: Joao Afonso Cardoso
• Diploma Examination: 11. March 2024
• Open Access: yes
• First Supervisor: Michael Wimmer
• Pages: 198
• Keywords: Animation, Limited animation, Line art, Image vectorization, Vector graphics, Deep learning, Machine learning

Abstract

Animation consists of sequentially showing multiple single frames with small mutual differences in order to achieve the visual effect of a moving scene. In limited animation, these frames are drawn as semantically meaningful vector images which could be referred to as clean animation frames. There are limited animation workflows in which these clean animation frames are only available in raster format, requiring laborious manual vectorization.This work explores the extent to which line-art image vectorization methods can be used to automatize this process. For this purpose, a line-art image vectorization method is designed by taking into account the structural information about clean animation frames. Together with existing state-of-the-art line-art image vectorization methods, this method is evaluated on a dataset consisting of clean animation frames. The reproducible evaluation shows that the performance of the developed method is remarkably stable across different input image resolution sizes and binarized or non-binarized versions of input images, even outperforming state-of-the-art methods at input images of the default clean animation frame resolution. Furthermore, it is up to 4.5 times faster than the second-fastest deep learning-based method. However, ultimately the evaluation shows that neither the developed method nor existing state-of-the-art methods can produce vector images that achieve both visual similarity and sufficiently semantically correct vector structures.

Additional Files and Images

Weblinks

• Entry in reposiTUm (TU Wien Publication Database)
• DOI: 10.34726/hss.2024.102471

BibTeX

@mastersthesis{metzger-2024-smv,
  title =      "Semantically meaningful vectorization of line art in drawn
               animation",
  author =     "Calvin Metzger",
  year =       "2024",
  abstract =   "Animation consists of sequentially showing multiple single
               frames with small mutual differences in order to achieve the
               visual effect of a moving scene. In limited animation, these
               frames are drawn as semantically meaningful vector images
               which could be referred to as clean animation frames. There
               are limited animation workflows in which these clean
               animation frames are only available in raster format,
               requiring laborious manual vectorization.This work explores
               the extent to which line-art image vectorization methods can
               be used to automatize this process. For this purpose, a
               line-art image vectorization method is designed by taking
               into account the structural information about clean
               animation frames. Together with existing state-of-the-art
               line-art image vectorization methods, this method is
               evaluated on a dataset consisting of clean animation frames.
               The reproducible evaluation shows that the performance of
               the developed method is remarkably stable across different
               input image resolution sizes and binarized or non-binarized
               versions of input images, even outperforming
               state-of-the-art methods at input images of the default
               clean animation frame resolution. Furthermore, it is up to
               4.5 times faster than the second-fastest deep learning-based
               method. However, ultimately the evaluation shows that
               neither the developed method nor existing state-of-the-art
               methods can produce vector images that achieve both visual
               similarity and sufficiently semantically correct vector
               structures.",
  month =      mar,
  pages =      "198",
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Research Unit of Computer Graphics, Institute of Visual
               Computing and Human-Centered Technology, Faculty of
               Informatics, TU Wien",
  keywords =   "Animation, Limited animation, Line art, Image vectorization,
               Vector graphics, Deep learning, Machine learning",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2024/metzger-2024-smv/",
}