Thesis defence : Audrey TROUILLOUD

The human visual system is particularly efficient at quickly understanding most of the visual environment despite its complexity. A first global and rudimentary representation of a visual scene would be possible with the rapid processing of the low spatial frequencies (LSF) of the visual signal. This first representation would also be used to very quickly generate predictions about the visual environment in order to anticipate and guide a more detailed analysis of the scene based on high spatial frequency (HSF) processing. Since LSF are mostly extracted in the peripheral retina, we hypothesized that predictions are mostly generated based on information available in peripheral vision. This led us in this doctoral thesis to compare visual processing in central and peripheral vision during scene categorization. To this end, we conducted several behavioral experiments in young adults with normal vision. First, we tested the categorization of hybrid images superimposing two different scenes, one in LSF and the other in HSF. We showed that the categorization of the HSF scene was better when the LSF scene was semantically congruent than incongruent, suggesting that LSF information, even when irrelevant for categorization, was automatically processed to guide HSF processing. This effect was stronger when the hybrid images were presented in peripheral vision than in central vision. We also tested the categorization of images revealing different parts of the same scene through circular rings of different retinal eccentricities. We observed that categorization performance was better for the most peripheral rings, suggesting that despite their low spatial resolution, the information available in peripheral vision is more useful than information available in central vision to categorize a scene very quickly. We then showed that the categorization of the central part of a scene was better when a semantically congruent peripheral ring was presented simultaneously than when an incongruent one was presented. Thus, the information available in peripheral vision would automatically guide the processing performed in central vision. Finally, we investigated the consequences of peripheral vision loss on scene categorization by conducting a behavioral experiment in patients with primary open-angle glaucoma (POAG). We tested hybrid image categorization in the patients' central vision and observed that the influence of LSF on HSF scene categorization decreased with the disease progression and progressive loss of stimulation in peripheral vision. All of this thesis work demonstrates the primordial role of peripheral vision in scene categorization and supports the predictive role of peripheral vision to guide the analysis of information present at the point of visual fixation.

Encadrante :
Directrice de thèse : Carole PEYRIN -  carole.peyrinuniv-grenoble-alpes.fr (carole[dot]peyrin[at]univ-grenoble-alpes[dot]fr)

Keywords: Central vision, Peripheral vision, Scene categorization, Spatial frequencies, predictive coding,

Read the thesis