Purchase this article with an account.
Po-Jang Hsieh, Edward Vul, Nancy Kanwisher; Top-down interpretation alters low-level visual processing. Journal of Vision 2009;9(8):858. doi: 10.1167/9.8.858.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
When looking at the classic RC James photograph for the first time, people can only identify the Dalmation after they are told that one exists in the image, or after it is explicitly outlined. In general, recognizing objects in such two-tone “Mooney” images is the prototypical example of the influence of global, top-down expectations on the interpretation of local, low-level image features - a process required for all image understanding. Here we used functional magnetic resonance imaging (fMRI) while subjects viewed two-tone images to investigate how global, top-down interpretation alters low-level visual processing. We compared the pattern of fMRI response while subjects viewed images in the following three conditions (occurring in this order): (1) viewing an ambiguous two-tone image when the object within it is not identified; (2) viewing the same image in grey-scale so that the object can be clearly identified; (3) viewing the original two-tone image again, but now the pictured object can be easily recognized due to the experience in (2). Our results show that BOLD response patterns in the early visual areas are more similar between conditions (2) and (3) than between (2) and (1). In other words, when participants know what objects are contained in an ambiguous two-tone image, the neural response to that image in early visual cortex becomes more similar to the neural response evoked by unambiguous the grey-scale photograph. The same effect is also observed in the lateral occipital complex. Thus, the high-level interpretation of a visual stimulus influences visual processing in early cortical areas. Our results suggest that the representation measured in low-level visual areas reflects a combination of high-level interpretation and low-level stimulus properties, as would be expected from a Bayesian inference framework.
This PDF is available to Subscribers Only