Effects of Transparency on Perceived Humanness: Implications for Rendering Skin Tones Using Optical See-Through Displays

IEEE Trans Vis Comput Graph. 2022 May;28(5):2179-2189. doi: 10.1109/TVCG.2022.3150521. Epub 2022 Apr 11.

Abstract

Current optical see-through displays in the field of augmented reality are limited in their ability to display colors with low lightness in the hue, saturation, lightness (HSL) color space, causing such colors to appear transparent. This hardware limitation may add unintended bias into scenarios with virtual humans. Humans have varying skin tones including HSL colors with low lightness. When virtual humans are displayed with optical see-through devices, people with low lightness skin tones may be displayed semi-transparently while those with high lightness skin tones will be displayed more opaquely. For example, a Black avatar may appear semi-transparent in the same scene as a White avatar who will appear more opaque. We present an exploratory user study ( N=160) investigating whether differing opacity levels result in dehumanizing avatar and human faces. Results support that dehumanization occurs as opacity decreases. This suggests that in similar lighting, low lightness skin tones (e.g., Black faces) will be viewed as less human than high lightness skin tones (e.g., White faces). Additionally, the perceived emotionality of virtual human faces also predicts perceived humanness. Angry faces were seen overall as less human, and at lower opacity levels happy faces were seen as more human. Our results suggest that additional research is needed to understand the effects and interactions of emotionality and opacity on dehumanization. Further, we provide evidence that unintentional racial bias may be added when developing for optical see-through devices using virtual humans. We highlight the potential bias and discuss implications and directions for future research.

Publication types

  • Research Support, U.S. Gov't, Non-P.H.S.
  • Research Support, Non-U.S. Gov't

MeSH terms

  • Computer Graphics*
  • Humans
  • Lighting
  • Skin Pigmentation*