Home > Research > Publications & Outputs > Investigating human-perceptual properties of "s...

Electronic data

  • 2020PowerPhD

    Final published version, 14.6 MB, PDF document

    Available under license: CC BY: Creative Commons Attribution 4.0 International License

Text available via DOI:

View graph of relations

Investigating human-perceptual properties of "shapes" using 3D shapes and 2D fonts

Research output: ThesisDoctoral Thesis

Published
  • Luther Power
Close
Publication date2020
Number of pages335
QualificationPhD
Awarding Institution
Supervisors/Advisors
Thesis sponsors
  • Engineering and Physical Sciences Research Council
Award date27/04/2020
Publisher
  • Lancaster University
<mark>Original language</mark>English

Abstract

Shapes are generally used to convey meaning. They are used in video games, films and other multimedia, in diverse ways. 3D shapes may be destined for virtual scenes or represent objects to be constructed in the real-world. Fonts add character to an otherwise plain block of text, allowing the writer to make important points more visually prominent or distinct from other text. They can indicate the structure of a document, at a glance.

Rather than studying shapes through traditional geometric shape descriptors, we provide alternative methods to describe and analyse shapes, from a lens of human perception. This is done via the concepts of Schelling Points and Image Specificity.

Schelling Points are choices people make when they aim to match with what they expect others to choose but cannot communicate with others to determine an answer. We study whole mesh selections in this setting, where Schelling Meshes are the most frequently selected shapes. The key idea behind image Specificity is that different images evoke different descriptions; but ‘Specific’ images yield more consistent descriptions than others. We apply Specificity to 2D fonts.

We show that each concept can be learned and predict them for fonts and 3D shapes, respectively, using a depth image-based convolutional neural network. Results are shown for a range of fonts and 3D shapes and we demonstrate that font Specificity and the Schelling meshes concept are useful for visualisation, clustering, and search applications. Overall, we find that each concept represents similarities between their respective type of shape, even when there are discontinuities between the shape geometries themselves. The ‘context’ of these similarities is in some kind of abstract or subjective meaning which is consistent among different people.