Abstract
A modulation of orientation in an orientationally narrow-band visual texture results in contrast modulations within two narrow orientation channels in the absence of a modulation in overall contrast. These contrast modulations typically do not occur in the channel which contains the bulk of the contrast in the texture but rather occur in ‘off-orientation’ channels. In other words, the relevant (contrast-modulated) information is contained within orientation channels distinct from channels containing irrelevant (unmodulated) information. As such, a perceptual mechanism which detects such texture modulations could potentially improve its performance by improving its ability to restrict second-order analysis to the relevant channels while ignoring the irrelevant channels. This might occur through a process of template retuning at the level of orientation-selective first-order perceptual filters. Whereas texture segmentation of orientation-modulated textures has indeed been shown to be subject to perceptual learning, it is not known whether this learning occurs through template retuning of orientation-selective first-order perceptual filters. We created textures containing contrast modulations in two distinct orientation channels. The contrast modulations in the two channels were either in-phase or counter-phase. The former results in textures which contain a modulation of overall contrast. These can be detected by mechanisms which integrate information across all first-order orientations and would hence not benefit by template retuning of their first-order filters. All textures also contained contrast in two separate orientation channels which were not modulated. The task of observers required detection of the contrast modulations in the relevant channels. Observers trained in at least ten sessions of 500 trials each. Performance showed a steady improvement across training sessions. This improvement was specific to retinal location but not to the first-order orientation channels which contained the relevant information. This suggests that perceptual learning of texture segmentation does not involve template retuning at the level of first-order perceptual filters.