Integration of elementary features in texture perception

点击翻译按钮获取中文摘要

Feature integration and its role in figure-ground segregation is a key problem of visual perception. According to established views, early retinotopic stages (V1, V2) serve local feature analysis, while shape detection and object recognition are located at higher ventral areas, which process objects in a feature and view independent manner. In line with an early channeling – later integration account strong cue combination effects (“synergy”) among orientation and spatial frequency were found. Texture shapes were much better detectable in Gabor random fields when there was feature contrast in both orientation and spatial frequency, however, only when they were barely visible by one feature alone and only when the Gabor elements formed a definite 2D shape. EEG recordings revealed a long-lasting negative potential starting at 130 ms specific for double-cue targets at electrodes around the inferior temporal region (TP), while there currently is no unique evidence for a double-cue correlate at central occipital (OZ) electrodes, suggesting feature integration at later shape-processing areas. Recently, the synergy effect was revisited using bandwidth modulated textures. Authors found that a low-level summary statistic, net contrast energy, computed from multiple scales and orientations, explains the synergy effect. The findings challenge a late, shape-based account of feature integration. In this project we readdress the origin of the feature synergy effect for the essential features of spatial luminance distributions in a Fourier sense (orientation, spatial scale and luminance) in texture perception. First, we try to establish that the strong cue combination effects in texture figure detection are fully captured by local energy computations. Second, we disentangle texture figure detection from shape discrimination with dedicated experimental designs, and explore whether the cue summation schemes found in shape identification tasks differ from the rules found in detection tasks. Third, EEG recordings will be analyzed by source modeling techniques to distinguish between activity originating in the primary visual cortex and the ventral processing stream. Applied with dedicated designs and tasks which disentangle detection and shape discrimination this will shed light on the time-course and neural correlates of cue summation in the two major processing regions for textures and shapes.