Anomalous visual experience is linked to perceptual uncertainty and visual imagery vividness

Abstract

An imbalance between top-down and bottom-up processing on perception (specifically, over-reliance on top-down processing) can lead to anomalous perception, such as illusions. One factor that may be involved in anomalous perception is visual mental imagery, which is the experience of "seeing" with the mind's eye. There are vast individual differences in self-reported imagery vividness, and more vivid imagery is linked to a more sensory-like experience. We, therefore, hypothesized that susceptibility to anomalous perception is linked to individual imagery vividness. To investigate this, we adopted a paradigm that is known to elicit the perception of faces in pure visual noise (pareidolia). In four experiments, we explored how imagery vividness contributes to this experience under different response instructions and environments. We found strong evidence that people with more vivid imagery were more likely to see faces in the noise, although removing suggestive instructions weakened this relationship. Analyses from the first two experiments led us to explore confidence as another factor in pareidolia proneness. We, therefore, modulated environment noise and added a confidence rating in a novel design. We found strong evidence that pareidolia proneness is correlated with uncertainty about real percepts. Decreasing perceptual ambiguity abolished the relationship between pareidolia proneness and both imagery vividness and confidence. The results cannot be explained by incidental face-like patterns in the noise, individual variations in response bias, perceptual sensitivity, subjective perceptual thresholds, viewing distance, testing environments, motivation, gender, or prosopagnosia. This indicates a critical role of mental imagery vividness and perceptual uncertainty in anomalous perceptual experience.

Fig. 1

a The ten faces viewed during training trials in all experiments, and in the main task of Exp. 2. b An example of a face embedded in noise from Face Set 1, used during training trials in Exp. 1. c An example of a face embedded in noise from Face Set 2, used during training trials in Exp. 1. d The checkerboard image that was used on attention control trials in Exp. 1

Fig. 2

A trend line (gray) and linear regression model (95% CI; gray shading) were fitted to the data for a Exp. 1a (N = 42); and b Exp. 1b (N = 37). Imagery vividness (mean rating on the VVIQ) is displayed on the x-axis and pareidolia proneness (binomial probability) is shown on the y-axis. Probability values were log-transformed for visualization purposes, so values that appear closer to 0 (log-transformed) are probabilities that are closer to 1

Fig. 3

a, b Yellow-to-red colors indicate z scores denoting positive correlations between subject responses and pixel grayscale values, and light blue-to-dark blue colors indicate z scores denoting negative correlations for Exp. 1a (a) and Exp. 1b (b). The range of z scores shown in the figure was set to match that of Rieth et al. To compare our results to those of Rieth and colleagues, we zoomed in on the central grid location where the authors of the previous study performed their analyses, and compared the (non-significant) clusters contained in our CIs to theirs (c). This panel is reproduced, on a Creative Commons license, from Fig. 3 in Rieth, C. A., Lee, K., Lui, J., Tian, J. & Huber, D. E. Faces in the Mist: Illusory Face and Letter Detection in i-Perception, dx.doi.org/10.1068/i0421, volume 2, pages 458–476, 2011, SAGE Publications Inc

Fig. 4

a A trend line (gray) and linear regression model (95% CI; gray shading) were fitted to the data of Exp. 2a (N = 25). The scatterplot shows the correlation between pareidolia proneness (number of misperceptions; y-axis) and imagery vividness (mean rating on the VVIQ; x-axis). b Two trend lines and linear regression models (95% CIs) were fitted to the data of Exp. 2a (N = 25) for correlations between pareidolia proneness (y-axis) and confidence (4 = high certainty; x-axis). The correlation with detection confidence is shown in grayscale, and the correlation with misperception confidence is shown in shades of red

Fig. 5

a A trend line (gray) and linear regression model (95% CI; gray shading) were fitted to the data of Exp. 2b (N = 23). The scatterplot shows the correlation between pareidolia proneness (number of misperceptions; y-axis) and imagery vividness (mean rating on the VVIQ; x-axis). b Two trend lines and linear regression models (95% CIs) were fitted to the data of Exp. 2b (N = 23) for correlations between pareidolia proneness (y-axis) and confidence (4 = high certainty; x-axis). The correlation with detection confidence is shown in grayscale, and the correlation with misperception confidence is shown in shades of red