Uncertainty and confidence in visual perception

Pascal Mamassian

CNRS & Université Paris Descartes, France

Visual perception is often seen as an inference problem where uncertainty comes from ambiguities in the world (e.g. the two 3D interpretations of the Necker cube), noise in the world (e.g. identifying a scene behind falling snowflakes), or noise in the visual system (e.g. synaptic noise). To deal with ambiguities, the visual system relies on prior knowledge such as the assumption that light comes from above our head. We have measured the characteristics of this assumption and found a systematic bias to the above-left for the illumination direction preference. In addition, we found evidence that the above-left assumption for the light source position is encoded early in the visual system and processed in a bottom-up way. Perceived uncertainty also determines the confidence in our perceptual decisions. With experiments on orientation discrimination, we show that objective confidence judgments are correlated with discrimination performance, and that these confidence judgements are not just based on some visibility cues in the stimulus. Overall, these results open the door for a thorough investigation of the processing of uncertainty and confidence by the visual system.