2011 年 82 巻 4 号 p. 385-391
Humans can recognize a complex natural scene even when it appears only briefly. The rapid recognition of natural scenes is accomplished by parallel processing of information based on multiple spatial frequencies and integration of this information. Previous studies have revealed the time course of integration of frequency-based information. However, it is still unclear how frequency-based information is integrated. There are two possible levels for the integration: One involves spatial integration of images and constructs a unified image, and the other entails semantic integration associated with the scene context level irrespective of spatial arrangements. We investigated the categorization accuracy of the low+high-pass images, in which a left-right mirror reversed low-pass image was superimposed on a nonreversed high-pass image or vice versa. In this context, the low+high-pass images were semantically integrable but spatially incongruent. The results indicated that accuracy of the low+high-pass images did not exceed the expected accuracy level estimated from separate presentations. This finding suggests that frequency-based information can be integrated spatially.