Abstract
The visual system is remarkably good at separating relevant object information from surrounding clutter or visual noise. Although attention is known to enhance sensory signals, the role of attention in extracting relevant information from noise is not well understood. Two prominent mechanisms have been proposed based on behavioral studies: Attention may act to amplify responses to all visual input, or it may act as a noise filter, reducing responses to irrelevant noise. A strict amplification mechanism should produce an attentional benefit only in low-noise situations, as by definition, such a mechanism would enhance both signal and noise. By contrast, a noise filtering mechanism should improve stimulus representations only at high noise levels. Here, we examined how attention modulates the representation of objects embedded in visual noise by using fMRI and multivoxel pattern classification to assess the discriminability of cortical responses to different types of objects. Observers either attended to or ignored images of objects belonging to one of four categories, embedded in various levels of visual noise. Pattern classification was used to measure the amount of object-category information present in individual visual areas. If attention acts to filter out irrelevant noise, then attending to the objects should lead to increased classification performance at high noise levels. Such effects were observed in all visual areas, including the primary visual cortex, suggesting that noise reduction begins to operate at the earliest stages of visual processing. However, if attention also acts as an amplification mechanism, then attending to the objects should increase classification performance at low noise levels as well. Such effects were only observed at higher stages of processing, including the FFA, PPA and lateral occipital complex. Taken together, these results suggest that visual attention improves our ability to discriminate objects by first de-noising visual input, and then amplifying this noise-reduced signal.
Meeting abstract presented at VSS 2012