Item

Abstract

Previous studies (Atick and Redlich, Field, Webster and Miyahara) have investigated how the visual system could optimally represent the 1/f amplitude spectrum of natural images. Computational
studies (Atick and Redlich, Field) suggest that the cortical representation ought to be
a “whitened” version of the amplitude spectrum of natural images, i.e. spatial frequencies are
equally represented despite the abundance of low spatial frequencies and dearth of high spatial
frequencies in photographs of real world scenes. Webster and Miyahara showed that adaptation
to natural images attenuates sensitivity to low spatial frequencies effectively supporting the
computational evidence. We attempt to measure to what degree different spatial frequencies
contribute to the percept of an image, in order to determine the extent of whitening of the input.
To do this we adapted subjects briefly (250ms) to textures (4 x 4 degrees) of different spatial
frequencies (1, 2, 4, 8, 16 cycles/degree, bandwidth 1.4 octaves—full width at half maximum).
Then we measured the perceived contrast of 1/f textures in the adapted region of the visual field
using the following procedure: After each interval of adaptation subjects judged whether the
texture in the adapted region had a higher or lower contrast than that of the same texture in a
non-adapted region. The contrast of the comparator texture (non-adapted) was changed after
each time the subject made a judgement according to a 1 up 1 down staircase. We found that
attenuation of perceived contrast, due to adaptation, is greatest when the adapting frequencies
are at the peak of the contrast sensitivity function. It seems there is some “whitening”; however
this is, at best, incomplete.