Abstract
The exact nature of the visual features that are brought to consciousness when one is engaging in mental imagery is still difficult to study empirically. The few studies that have attempted to reconstruct mental images obtained poor quality results due in part to the extremely sparse coverage of the “scene space” (Shen et al., 2019). Hence, the aim of the present study was to reconstruct better quality mental images by increasing the sampling resolution of the visual features within the scene space using the Bubbles method (Gosselin & Schyns, 2001) and electroencephalography (EEG). So far, we have recorded the brain activity of four participants during two alternating tasks divided into 6 one-hour sessions. In the perception task, participants were presented with two scene images through different sets of randomly located Gaussian apertures or “bubble masks” (1,500 trials per image in total). In the mental imagery task, subjects were shown the two stimuli successively and asked to imagine either the first, or second one (300 trials per image in total). For each participant and for each scene, we correlated the EEG activity patterns between mental imagery and visual perception. Specifically, we correlated the EEG activity in each trial of the visual imagery task with the EEG activity elicited from partial presentation of the scenes through bubbles. Correlation-weighted sums of the associated bubble masks were computed in order to produce, for the first time, “classification images” of mental images.