Leeds Beckett University
Browse

Development of Digital Art Techniques to Test Perception of Photorealistic Computer Generated Characters

thesis
posted on 2023-03-02, 14:43 authored by Darren WallDarren Wall

This thesis reviews and discusses the creation of photorealistic virtual human faces and tests audience perception of a series of specifically developed visual assets. This provides a detailed look at the human perception of such visuals. In addition, specific techniques required to produce and render the photorealistic visuals are investigated. Viewer testing informs the development of a series of computer-generated (CG) animated assets, through which the thesis introduces a range of metrics that shape the viewers’ experience. Moreover, technical and creative recommendations to the artist are provided in relation to developing similar assets. 

The literature review considers how visuals of a realistic, although non-real, nature can in certain circumstances elicit a negative response in the viewer. This response can manifest as a feeling of repulsion, rather than empathy, towards the CG subject. Mori’s (1970) theory of the “uncanny valley effect” is considered as a hypothesis for predicting this response. This negative response is problematic for an artist attempting to develop photorealistic visuals. The uncanny valley effect is looked at in detail and established work in the field is built upon to consider animated visuals in addition to static images. Previous research does not fully address the perception of animated characters. This thesis shows how the addition of motion may affect viewer perception of the visuals. Moreover, the impact of image composition is taken into account. 

Optimal asset generation processes are determined, employing relevant technologies, artistry and techniques. This was steered by experience and perception of the visuals. The methodology employed saw appropriately targeted individuals complete questionnaires and self-report documents to identify their objective feelings towards sample visual media. Test data was gathered, analysed and used to identify key visual triggers in the imagery. Accrued data informed the approach to develop refined CG assets, which were again tested. 

Test One, Real or CG, gauges the accuracy of human perception in determining real (photographic) imagery from (CG) imagery. This provides a more comprehensive understanding of how and why viewers’ decisions and opinions were made. The test also aids the development of future research in identifying the key elements that inform the test subject's decisions. 

Test Two, Is it in The Detail?, establishes whether a viewer requires an entire image or a fragment of an image to determine whether an image is real or CG. Here an informed decision is reached about whether the whole picture or merely close-ups of the key features delivers greater accuracy. Additional emphasises is given to the rationale for these differences. 

Test Three, Photorealistic Character, takes the earlier tests a stage further, with the addition of movement. Participants judged a digital human asset in terms of realism in its static form and then again while in motion. The results of each exposure to the visuals were then compared. An additional comparison between full frame and close-up sections of animation was also conducted. Masahiro Mori’s (1970) uncanny valley theory regarding differences between perceptions of static and moving virtual humans is discussed here in relation to the recorded data. 

The thesis shows that movement and composition of visuals have an impact on viewer perception of CG humans. The test data also confirms that the accuracy of the motion impacts upon the perceived human-likeness of the visuals.

History

Qualification name

  • PhD

Supervisor

Cope, Nick ; Behringer, Reinhold ; Bryant, Anthony

Awarding Institution

Leeds Beckett University

Completion Date

2017-11-01

Qualification level

  • Doctoral

Language

  • eng

Publisher

Leeds Beckett University

Usage metrics

    LBU Theses and Dissertations

    Categories

    No categories selected

    Licence

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC