Ensemble statistics of faces

• Main Author: Ying, Haojiang
• Other Authors: Xu Hong
• Format: Theses and Dissertations
• Language: English
• Published: 2019
• Subjects: DRNTU::Social sciences::Psychology::Experimental psychology
• Online Access: https://hdl.handle.net/10356/89893; http://hdl.handle.net/10220/47740
• Institution: Nanyang Technological University

Ensemble statistics describes the ability of the visual system to summarize the information provided by a group of objects. This thesis refines the theoretical framework of ensemble statistics of faces throughout four studies. Study 1 investigated involuntary ensemble statistics of facial expressions during rapid serial visual presentation (RSVP). Prolonged exposure to RSVP of faces led to significant facial expression adaptation aftereffects, which could be explained by ensemble statistics. Further testing clarified that this representation was only modulated by the mean information in that stream. In summary, Study 1 examined the involuntary ensemble coding of facial expressions and hinted at the potential mechanism behind the formation of face space. Comparing adaptation aftereffects, Study 2 showed that temporal and spatial ensemble statistics of faces arise from distinct mechanisms that produce qualitatively different perceptual outcomes. The visual system extracts the low-level ‘computational’ average from faces when faces are presented individually across time. However, the spatial ensemble statistics summarizes the higher-level gist. Study 2, for the first time, showed there are distinctive mechanisms for ensemble coding of the same facial charateristics. Studying facial attractiveness adaptation and the ‘cheerleader effect’ (i.e. faces are perceived as more attractive when surrounded by others rather than being alone), Study 3 linked these two important phenomena in facial attractiveness with ensemble statistics. The mean attractiveness of the crowd (determined by ensemble statistics) biased the perceived attractiveness of the subsequently viewed face (the adaptation aftereffect). Similarly, the levels of attractiveness in a simultaneously presented crowd could also affect a target face (the ‘cheerleader effect’). Both past and present experience (determined by ensemble statistics) therefore impact the perception of facial attractiveness. Study 3 showed that the ensemble coding is ubiquitous in face perception and shapes two important phenomena in facial attractiveness. Also, the findings suggested how external factors, the previous and present exposures to faces, affect the attractiveness perception. Converging evidence from Study 4 emphasized the role of attention in both explicit and implicit ensemble statistics of facial expressions, and also suggested that the ensemble statistics is a weighted average of the visual input rather than the simple average. This study further clarified the averaging mechanism of ensemble coding and unveiled the relationship between ensemble coding of the face and attention. Taken together, the thesis examined the mechanisms involved in high-level ensemble statistics and the relationship among ensemble statistics, face perception, and attention. The results may shed light on a more comprehensive understanding of face perception and visual processing.