Abstract
Since natural visual images rarely contain locally unambiguous information about visual objects and their boundaries, the visual system must use prior knowledge of relevant objects to constrain possible interpretations of a given image. We explored the neural mechanisms of this process using camouflaged objects, which are difficult to detect without prior knowledge, even when in ‘plain view’. We created visual scenes each of which contained a single novel foreground object (‘digital embryo’, Brady and Kersten, 2003) camouflaged against a cluttered background of additional novel digital embryos. Subjects were trained to detect five different individual foreground embryos (‘learned targets’) using a bootstrapped learning paradigm. All subjects performed at chance levels during initial phases of the training, indicating that the camouflaged embryos were not detectable without prior knowledge. However, 5 of the 7 subjects learned within 300 trials to reliably detect the designated targets (d' analysis, p < 0.05). Following successful training, rapid event-related fMRI scans were carried out while the subject reported whether or not the scene presented during a given trial contained a learned target. We found several cortical foci in either hemisphere that were significantly more responsive during trials that contained a learned target than during trials that contained an unfamiliar target. Furthermore, some foci, including the lateral occipital complex (LOC), were selectively responsive when the subject correctly identified a learned target (p < 0.05). Together, these foci may represent parts of a neural mechanism by which prior knowledge of visual objects is brought to bear on visual perception.
This work was supported by NEI grant R01 EY015261 and ONR grant N00014-05-1-0124 to DK. The 3T scanner at the University of Minnesota Center for Magnetic Resonance Research is supported by BTRR P41 008079 and by the MIND Institute.