Evaluating the perception of virtual characters by manipulating components of their physical appearance

Abstract

This work explores the possibilities of studying the perception of 3D generated (virtual) based on the knowledge of perceptual organisation and methods used to study human behaviour. Our main intention is to present studies which explore responses towards virtual characters by manipulating specific components of their appearance. The literature on virtual characters presents some examples on how appearance could affect the perceiver's response. The most well-known phenomenon is the so called ?Uncanny Valley? which describes the loss in appeal and increase in eeriness when the virtual character reaches a certain level of detail, closely resembling a human being. The research has since shown that the concept of uncanny valley is not as clear as proposed by Mori (1970) and some methodological and conceptual discrepancies have been revealed. In addition, disinterest and lack of appeal can be found in the responses to non-realistic characters as well. While virtual characters are becoming increasingly more prevalent in movies, gaming industry, as well as being used in education, therapy and are expanding to other areas of life, a better understanding of how they are perceived is of crucial importance. Our work presents experiments which were designed using virtual characters in styles created by manipulating material properties, model geometry and rendering effects. For animated characters, natural human motion recorded with a motion capture system was used. In the first study, we focused on the perception of gender from motion cues, by adding the expression of emotion and analysing the possible effect of stereotypes on how gender is perceived. For the second study, we assessed combinations of different realism levels of shape and materials in an attempt to find the least and most appealing results and find how this manipulation affects perception of emotion. The last two studies were built in interactive and immersive environments where we used proximity in combination with subjective reports to explore the effects of agency and personality on peoples' attitudes towards virtual characters of different visual styles. Our last study is particularly important since it was conducted on an extremely large sample size (1106 participants) and used a between-group design to study the effect of appearance on people's perception. Our findings show that overall, appearance is an important factor which does not only affect appeal but can change the way perceivers interpret the character's behaviour, empathise with it, and behave around it. We find indications that realism is a positive component of appearance which may even be preferred to stylisation, while certain manipulations of realism in graphics elements will result in unappealing combinations. We also show that indirect measures are a good addition to the subjective reports, although these measures and the outcomes of using them need to be investigated further. We are confident that our work is an important contribution to the theoretical knowledge on perception in computer graphics, as well as an applicable value in the form of guidelines for character design.