Vision and imagery: The role of cortical attractor dynamics

Mental imagery presents individual differences and degrees of vividness which appear discrete. We suggest that visual brain areas follow an attractor dynamics giving them two properties: pattern completion and bistability. Bistability implies that visual brain areas are entirely active or inactive. Vision and imagery use the same system, vision being bottom-up and imagery top-down input. In imagery, different configurations of active visual areas account for different imagery types. The more vivid the imagery, the more occipital areas are recruited. The dissertation comports three parts. The first part details this theory. The second part presents a computer model where a layer of neurons represent an image by wavelet (Gabor) components. Lateral connections make that layer an attractor network. The properties observed support the statements made above about attractor networks: The model is bistable, has an activation threshold, can improve the shape of a noised pattern (restoration by pattern completion), and can combine top-down and bottom-up sub-thresholds inputs to reach activation. The third part is a psychological experiment testing this prediction on how imagery interacts with vision: when imagery is going on, low-level visual processes are more efficient if the visual image is similar to the mental image. 63 subjects completed 160 trials each. In 80 trials, they are cued by a visualization, and must afterwards make a visual judgement on a clock image: they decide if both hands of the clock are in a shaded plane which covers half of the image. When visualizing the right clock, subjects are faster than when visualizing the wrong clock. A control condition (80 trials) used a visual rather than imaged cue (Previous studies lacked controls for acquisition strategies, such as eye movement planning or attention allocation). The effect is smaller with the visual cue, and restricted to a particular sub-case where one expects acquisition strategies. This experiment demonstrates that the influence of imagery on perception depends on the content of the image. It is an interaction between mental images and visual images, rather than an interaction between "systems". Besides this main finding, lateralization results confirm the growing hypothesis that composite imagery is strongly lateralized.