Scene categorization after priming/adaptation to amplitude statistics

Abstract

The visual system is able to rapidly and accurately categorize natural scenes. Whilst diagnostic information for categorization is considered to be contained mainly in the localized phase structure of an image, there is some debate as to the potential role of amplitude spectra information. For priming (Guyader et al., 2004) and adaptation (Kaping et al., 2007) studies, the presentation of amplitude information prior to image categorization has been shown to shift observer judgements, suggesting an influence for amplitude statistics in image categorization. We created prime/adaptation stimuli by averaging the amplitude spectra of a large number of natural or man-made stimuli. These images (with randomized phase information) were presented for durations ranging between 50 ms and 4 s, prior to the brief presentation of a test stimulus (14 ms). Test stimuli consisted of averaged amplitude spectra of a pair of images, one natural and one man-made. Phase information for a test image was a blend between the two single images, from 20% natural, 80% man-made to 80% natural, 20% man-made. Observers classified each test image as natural or man-made. Our observers showed a shift in their natural/man-made category boundary following longer presentations of the adaptation stimulus, yet limited priming effects. These results indicate that amplitude information may be of limited importance to image classification.