Texture segmentation in human perception: a combined modeling and fMRI study

The human visual system uses texture information to segment visual scenes into figure and ground. We developed a computational model of human texture processing [Thielscher A, Neumann H (2003) Neural mechanisms of cortico-cortical interaction in texture boundary detection: a modeling approach. Neuroscience 122:921-939] which allows us to examine the functional roles of early and intermediate stages of the ventral visual pathway in figure-ground segmentation. In particular, the model highlights the central role of cells in mid-level areas (such as V4) with larger receptive fields in the robust identification of texture boundaries and pop-out stimuli even under noisy conditions. A straightforward prediction of the model is that the activity of cells in mid-level, but not early visual areas directly co-varies with the saliency of the texture borders in the visual scene. Consequently, their activity should directly correlate with the saliency of pop-out texture regions as accessed in psychophysical studies [Nothdurft HC (1991) Texture segmentation and pop-out from orientation contrast. Vision Res 31:1073-1078]. This prediction explicitly derived from the model was tested using functional magnetic resonance imaging. The saliency of texture bars composed of oriented line items was varied by parametrically changing the defining orientation contrast between fore- and background lines. Consistent with the model, increasing contrast at texture boundaries resulted in a monotonic increase of blood oxygen level dependent responses in mid-level, but not early visual areas. Our modeling and imaging results indicate that mid-level visual areas form a key stage in figure-ground segregation by gradually signaling the salience of region boundaries defined by orientation contrast.