Models of Early Spatial Vision
Most current models of early spatial vision assume independent spatial-frequency-and orientation-selective channels as an initial stage of the image encoding process, typically in form of a Gabor-filter bank (loosely a \"wavelet\" analysis of the input image). In my talk I will briefly review some of the evidence for this assumption as well as extraneous factors that may help its popularity, such as the neurophysiology of simple cell receptive fields and notions of optimality in terms of joint space/spatial-frequency resolution and natural image encoding. In the second part of my talk I will focus on recent experiments that have used pulse trains, plaids and amplitude modulated gratings as stimuli. The results of these experiments pose serious problems for current models of early vision that seek to explain human spatial vision in terms of the outputs of filters or channels tuned to narrow bands of spatial frequency, even when the models incorporate significant non-linearities.