Thesis defence : Pauline ROSSEL

 

Current models of visual perception consider perception as a proactive process. In this context, perceiving would consist of permanently matching the characteristics of a visual stimulus to expectations or predictions about it, developed on the basis of previous experiences and learned regularities in the environment. However, the mechanisms by which these expectations (or predictions) influence information processing and ultimately shape the subjective experience of visual stimuli remain debated. While some studies suggest that perception is biased in favour of what can be predicted and expected in the visual environment - leading more often than not to a reliable perception of the latter - other studies suggest, on the contrary, that perception is dominated by what is unexpected - and therefore more informative. The aim of this thesis was to determine to what extent and in what way the subjective perception of complex visual stimuli such as scenes and objects is likely to be modified as a function of their expected or unexpected character, as well as of different visual constraints.To do this, we conducted 6 psychophysical experiments. The first three experiments enabled us to demonstrate that predictions have a qualitative influence on perception. More specifically, we showed that blurred scenes and objects are subjectively perceived as sharper when they can be predicted on the basis of contextual information available in the visual pattern than when this contextual information is absent, even though these stimuli are physically the same in both conditions. The results of the fourth experiment also showed that the perceptual effects of the predictions varied according to the quality of the visual signals: these effects were all the stronger when the visual stimuli were noisy (i.e. blurred). In the fifth experiment, we aimed more specifically to study how these perceptual effects are influenced by the validity of the predictions (correct or incorrect predictions). We first observed that when the visual signal is noisy (e.g. very blurred), expected objects (presented in a congruent context leading to correct predictions) are perceived as sharper than the same unexpected objects (presented in an incongruent context leading to incorrect predictions). The validity of the predictions would considerably help the perception of noisy stimuli. On the other hand, when the visual signal was less noisy (for example, relatively sharp objects), unexpected objects were perceived as sharper than expected objects. In Experiment 6, we wanted to test whether these effects also depended on temporal constraints, but our results did not show any modulation of the effect of the predictions according to the time at which the stimuli were presented. Overall, the results of this work make it possible to clarify the mechanisms by which predictions qualitatively influence perception by highlighting the role of the validity of the predictions (expected vs. unexpected stimulus), as well as the role of the reliability of the visual signal (for example, the level of blur). We have interpreted them in the context of recent theoretical models according to which predictive mechanisms make it possible to optimise the processing of visual information so as to construct percepts that are both reliable and informative.