Information

  • Publication Type: Master Thesis
  • Workgroup(s)/Project(s):
  • Date: May 2019
  • Date (Start): 1. July 2017
  • Date (End): 27. May 2019
  • Diploma Examination: 27. May 2019
  • Open Access: yes
  • First Supervisor: Ivan ViolaORCID iD
  • Pages: 84
  • Keywords: visualization

Abstract

Advanced rendering algorithms such as suggestive contours are able to depict objects in the style of line drawings with various levels of detail. How to select an appropriate level of detail is based on visual aesthetics rather than on substantial characteristics like the accuracy of 3D shape perception. The aim of this thesis is to develop a novel approach for effectively generating line drawings in the style of suggestive contours that are optimized for human 3D shape perception while retaining the amount of ink to a minimum. The proposed post-processing meta-heuristic for optimizing line drawings uses empirical thresholds based on probing human shape perception. The heuristic can also be used to optimize line drawings in terms of other visual characteristics, e.g., cognitive load, and for other line drawings styles such as ridges and valleys. The optimization routine is based on a conducted perceptual user study using the gauge figure task to collect more than 17, 000 high-quality user estimates of surface normals from suggestive contours renderings. By analyzing these data points, more in-depth understanding of how humans perceive 3D shape from line drawings is gained. Particularly the accuracy of 3D shape perception and shape ambiguity in regards to changing the level of detail and type of object presented is investigated. In addition, the collected data points are used to calculate two pixel-based perceptual characteristics: the optimal size of a local neighborhood area to estimate 3D shape from and the optimal local ink percentage in this area. In the analysis, a neighborhood size of 36 pixels with an optimal ink percentage of 17.3% could be identified. These thresholds are used to optimize suggestive contours renderings in a post-processing stage using a greedy nearest neighbor optimization scheme. The proposed meta-heuristic procedure yields visually convincing results where each pixel value is close to the identified thresholds. In terms of practical application, the optimization scheme can be used in areas where high 3D shape understanding is essential such as furniture manuals or architectural renderings. Both the empirical results regarding shape understanding as well as the practical applications of the thesis’s results form the basis to optimize other line drawing methods and to understand better how humans perceive shape from lines.

Additional Files and Images

Additional images and videos

Additional files

Weblinks

BibTeX

@mastersthesis{plank-2017-sldg,
  title =      "Effective Line Drawing Generation",
  author =     "Pascal Plank",
  year =       "2019",
  abstract =   "Advanced rendering algorithms such as suggestive contours
               are able to depict objects in the style of line drawings
               with various levels of detail. How to select an appropriate
               level of detail is based on visual aesthetics rather than on
               substantial characteristics like the accuracy of 3D shape
               perception. The aim of this thesis is to develop a novel
               approach for effectively generating line drawings in the
               style of suggestive contours that are optimized for human 3D
               shape perception while retaining the amount of ink to a
               minimum. The proposed post-processing meta-heuristic for
               optimizing line drawings uses empirical thresholds based on
               probing human shape perception. The heuristic can also be
               used to optimize line drawings in terms of other visual
               characteristics, e.g., cognitive load, and for other line
               drawings styles such as ridges and valleys. The optimization
               routine is based on a conducted perceptual user study using
               the gauge figure task to collect more than 17, 000
               high-quality user estimates of surface normals from
               suggestive contours renderings. By analyzing these data
               points, more in-depth understanding of how humans perceive
               3D shape from line drawings is gained. Particularly the
               accuracy of 3D shape perception and shape ambiguity in
               regards to changing the level of detail and type of object
               presented is investigated. In addition, the collected data
               points are used to calculate two pixel-based perceptual
               characteristics: the optimal size of a local neighborhood
               area to estimate 3D shape from and the optimal local ink
               percentage in this area. In the analysis, a neighborhood
               size of 36 pixels with an optimal ink percentage of 17.3%
               could be identified. These thresholds are used to optimize
               suggestive contours renderings in a post-processing stage
               using a greedy nearest neighbor optimization scheme. The
               proposed meta-heuristic procedure yields visually convincing
               results where each pixel value is close to the identified
               thresholds. In terms of practical application, the
               optimization scheme can be used in areas where high 3D shape
               understanding is essential such as furniture manuals or
               architectural renderings. Both the empirical results
               regarding shape understanding as well as the practical
               applications of the thesis’s results form the basis to
               optimize other line drawing methods and to understand better
               how humans perceive shape from lines.",
  month =      may,
  pages =      "84",
  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 =   "visualization",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2019/plank-2017-sldg/",
}