The Neurological Traces of Look-Alike Avatars

Abstract

We designed an observational study where participants (n = 17) were exposed to pictures and look-alike avatars pictures of themselves, a familiar friend or an unfamiliar person. By measuring participants' brain activity with electroencephalography (EEG), we found face-recognition event related potentials (ERPs) in the visual cortex, around 200-250 ms, to be prominent for the different familiarity levels. A less positive component was found for self-recognized pictures (P200) than pictures of others, showing similar effects in both real faces and look-alike avatars. A rapid adaptation in the same component was found when comparing the neural processing of avatar faces vs. real faces, as if avatars in general were assimilated as real face representations over time. ERP results also showed that in the case of the self-avatar, the P200 component correlated with more complex conscious encodings of self-representation, i.e., the difference in voltage in the P200 between the self-avatar and the self-picture was reduced in participants that felt the avatar looked like them. This study is put into context within the literature of self-recognition and face recognition in the visual cortex. Additionally, the implications of these results on look-alike avatars are discussed both for future virtual reality (VR) and neuroscience studies.

Keywords: avatars; event related potentials; face-recognition; memory; self-recognition; visual cortex.

Figure 1

(A) Creation of the look-alike avatar, the three pictures used for the avatar generation on the top. On the bottom, the final pictures used for the experiment as real and virtual. (B) Experimental execution. The six faces (self-real, self-virtual, familiar-real, familiar-virtual, unfamiliar-real, unfamiliar-virtual) were randomly ordered in blocks of 10. Each face was displayed for 300 ms, followed by a variable time of 740–1340 ms in which a fixation cross appeared. After each block there was a short resting period of 2 s for blinking.

Figure 2

Boxplots of the scores for the different faces, from the pre and post questionnaires. The scores range from 1 (not alike at all) to 5 (totally looks alike). Participants rated the faces by (A) Realism: they compared the pictures to the real person that they knew; (B) Similarity: they compared the real picture to the picture of the avatar face independently of whether they knew not the person in the picture. The thick horizontal bars are the medians and the boxes are the interquartile ranges. The whiskers range from max (lower quartile − 1.5 * IQR, smallest value) to min (upper quartile + 1.5 * IQR, largest value). Values outside of this range are marked by *.

Figure 3

Grand average event related potentials (ERPs) for the 17 subjects of the parietal electrodes (PO7, P7) and (PO8, P8) elicited during the first 100 trials by (A) Real faces: self, familiar, and other. (B) Virtual faces: self, familiar, and unfamiliar. The current source density (CSD) topographical plots show how the difference between others-self for the P200 component is mainly located in the occipito-parietal cortex. A low pass filter (15 Hz, half-amplitude cut-off) was applied in these grand averaged graphs.

Figure 4

Grand average ERPs of the 17 subjects in the parietal electrodes (PO7, P7) and (PO8, P8) elicited by the real and the computer generated faces for the first 100 trials. There is a significant difference in the P200, mainly in the parietal left electrodes. A low pass filter (15 Hz, half-amplitude cut-off) was applied in these grand averaged graphs.

Figure 5

The top panel shows the time evolution of the P200 amplitude in the left tempo-parietal cortex (P07, P7); presenting the cumulative voltage over blocks of 30 trials (the error bars show the standard error of the different participants ERPs). We observe how in the first trials the virtual and the real faces are processed as different objects; however, this effect is reduced after the overexposure. In the panel below we observe the grand averaged ERPs of the first and last 50 trials, and a clear reduction of the P200 component is also observed. The topographical plots show the CSD of the P200 component difference (avatar-real) in the scalp, we can see how the difference decreases over the last trials. A low pass filter (15 Hz, half-amplitude cut-off) was applied to these grand averaged graphs.

Figure 6

Correlation between the virtual self-face identification scores in the realism question and the P200 voltage difference between self-avatar–self-real. The closer the voltage difference is to zero the greater the identification.