Abstract
Computational models can effectively categorize scene images based solely on their Fourier amplitude spectra “signature profiles” (Oliva & Torralba, 2001). However, recent behavioral studies have shown that amplitude spectra signatures are insufficient for scene category recognition, suggesting instead that characteristic phase spectra signatures are necessary. Using a scene categorization masking paradigm, Loschky et al. (2007) demonstrated that phase-randomized scenes-as-masks produce weaker masking than normal-scenes-as-masks, and that phase-randomized masks do not produce category-specific masking effects. Nevertheless, because phase randomization renders a scene-as-mask unrecognizable, the loss of masking could reflect a reduction in conceptual masking rather than a loss of local phase structure. Here, we tested the hypothesis that local, unrecognizable, category-specific features in the phase spectrum of a mask are sufficient to produce category-specific masking effects. Composite phase spectrum masks (CPSMs) were created by sampling different portions of the phase spectra of images within a single scene category (beach, forest, home interior, store interior, street, or tall building), thus CPSMs were naturally structured, but were not recognizable as members of their categories. Using a “go no-go” paradigm, images from pairs of categories were masked by CPSMs from each category. The results showed that when the basic-level category of the target and mask matched, masking was weaker than when they differed, thereby supporting the hypothesis that low-level category-specific features do mask specific image categories. Interestingly, regardless of CPSM category, there was a separate effect of whether the target-distracter image pairs were from the same or different superordinate categories (natural vs. man-made, or indoor vs. outdoor). Scene categorization at the basic level was easier when the pair of basic level categories came from different superordinate categories, consistent with the claim that scene category distinctions are first made at the superordinate level, and only later at the basic level (Loschky & Larson, 2008).