Abstract
Contemporary theories of object recognition posit that an object's position in the visual field is quickly discarded at an early stage in visual processing, in favor of a high level, position-invariant representation. The present study investigated this supposition by examining how the location of an object is encoded in the brain as a function of time. In three experiments, participants viewed images of objects while brain activity was recorded using MEG. In each trial, subjects fixated a central point and images of objects were presented to variable locations in the visual field. The nature of the representation of an object's position was investigated by training a linear classifier to decode the position of the object based on recorded physiological responses. Performance of the classifier was evaluated as a function of time by training the classifier with data from a sliding 10ms time window. The classifier's performance for decoding the position of the object rose to above chance levels at roughly 75ms, peaked at approximately 115ms, and decayed slowly as a function of time up to 1000ms post-stimulus onset. Within the interval of 75 to 1000ms, classification performance correlated with the angular distance between targets, indicating a metric representation of visual space. Notably, prior to the time that classification performance returned to chance, object category information could be decoded from physiological responses; and, participants were able to accurately make high level judgments about the objects (i.e. category and gender for faces). These findings suggest that position may be a fundamental feature encoded in the representation of an object, in contrast to the notion that position information is discarded at an early stage of visual processing.