Seminar Talk by Funda Yıldırım
On the contribution of contour mechanisms in early visual cortex to shape perception: an comparative study of neural networks and statistical modeling
The visual system tends to group clusters of similar parts into a single shape. The perceived shape of such an object – consisting of many similar small parts – can be manipulated by creating a perceptual conflict between the orientation and the position of the parts. This has the interesting consequence that the position of parts and intermediate illusory contours can appear displaced compared to their physical position. Here, we used this shape-dependent change in perceived position as a pointer to identify visual regions involved in shape perception. We hypothesized that if the percept of a shape is primarily evoked bottom-up through (illusory) contours generated at an early level in visual cortex, it should be possible to read-out the displaced contour activity in V1 and V2, as these areas have previously been associated with contour integration. On the other hand, if it is the perceived shape that drives the perceived positional change of the parts and contours, activity in early visual cortex should reflect the physical position of the parts and not differ depending on the perceived shape. We present the results obtained via recurrent neural networks and population receptive field models and discuss their congruency. Last but not least, we use this case to argue how machine learning can be utilized to shed light on cognitive neuroscience studies.