Objects in Scenes - How scene structure shapes visual object representations

点击翻译按钮获取中文摘要

In everyday situations, our visual environments are highly complex, consisting of a multitude of different objects. These objects do not appear in unpredictable locations in space, but adhere to typical, frequently encountered positional structures. For example, lamps are commonly hanging from the ceiling, while we expect to see carpets on the floor. Does the massive experience with objects seen in such highly predictable locations influence the way in which objects are processed in the visual system? Here, I propose that recurring regularities in object positions shape visual processing channels that are optimally tuned for efficiently coding specific objects appearing in specific parts of the visual field. In this project, I will use behavioral experiments and a combination of fMRI and EEG recordings to test two key predictions derived from this proposal: (1) Object information is non-uniformly distributed across the visual field, with objects being coded more efficiently when they appear in their typical locations. As a consequence, typically positioned objects (e.g., a lamp in the upper visual field) should be perceived more accurately and their representations should be better discriminable on a neural level - as compared to atypically positioned objects (e.g., a lamp in the lower visual field). (2) The similarity of individual object representations in visual cortex is partly determined by real-world structure, so that objects associated with similar locations are coded similarly. Hence, representational similarity in visual cortex should be explicable by typical object-to-object distances (e.g., a lamp should be more similarly represented to a table than to a carpet), even when the objects are presented centrally and in isolation. The idea of location-specific processing channels along the visual hierarchy suggests that the effects of typical spatial structure can be observed in object-selective visual cortex around the time of initial object processing. The combination of fMRI and EEG allows for testing this prediction, as it can precisely reveal how the spatiotemporal dynamics of object vision are related to the structural properties of our everyday environments. The results of the project can indicate that visual object representations can only be fully understood in the context of real-world environments and their positional structure. Furthermore, these novel insights can provide an explanation for the efficient perception of natural scenes, which contain a large amount of different - but regularly structured - objects.