Abstract
We investigated the role of feedback in texture segmentation. We tested the texture perception of a patient (HJA) with visual agnosia who has a large bilateral lesion of the ventral cortex, likely to include V2, V3 and V4. Previously he has been found to have difficultly segmenting overlapping figures and grouping object parts. Models of texture segmentation (e.g. Theilscher & Neuman (2003) Neuroscience 122 p921–939) propose that simple segmentation can be done by early visual areas but feedback from higher visual areas (such as V4) is required for more difficult tasks.
Method: Participants viewed a 12×12 grid of lines and detected the presence of a 7×2 shape defined by a difference in orientation between these lines and the vertical background lines. The orientation contrast (difference) between the lines in the shape and background lines was measured at different levels of background orientation noise.
Results: Control participants easily detected the shape in aligned background element but needed more orientation contrast with there was orientation noise in the background. HJA was poor with all backgrounds, particularly so with aligned background elements. Adding additional random jitter to the background elements improved HJA's performance, but not significantly. Changing the orientation of the background elements to −45° and halving the density of the line elements improved his performance. Doubling the density improved performance only when there was no background orientation noise, performance was worse when background noise was increased.
Conclusions: HJA performs poorly when background elements are easier to group or align. His performance is improved when grouping is reduced. V1 appears to be insufficient to perform texture segmentation when the background of the display is homogeneous. Texture segmentation must, therefore, rely on information coming from, or through, ventral extra-striate areas to suppress the effects of alignment in early visual areas.