Jonathan VACHER (Ecole Normale Supérieure de Paris) has given a talk on Zoom on October 11th.
https://jonathanvacher.github.io/
Short abstract:
A necessary condition to make progress in our understanding of visual perception is to properly understand what is perceived ie natural images. While natural images still lack of interpretable generative models, natural textures are well-modelled and understood. Gaussian textures (characterized by the power spectrum) and naturalisitic textures (characterized by wavelet higher-order statistics) have been instrumental in understanding aspects of visual perception and of neural coding. Here, I will introduce these texture synthesis frameworks and I will show how improving them can lead to tackle new questions about the dynamical aspects of perception and about the mid-level perception. Finally, I will show how designing these texture models led me to question the coherence of the two key psychometric protocols I've used.