Proceedings of The Physiological Society
University of Cambridge (2004) J Physiol 555P, C163
Contrast masking and facilitation in human psychophysical experiments using natural scene stimuli
M. Chirimuuta and D.J. Tolhurst
Department of Physiology, University of Cambridge, Cambridge CB2 3EG, UK
It has repeatedly been observed (e.g. Foley 1994) that the minimum detectable contrast difference of a sinusoidal grating is a function of the base contrast of the grating. For low contrast gratings, the discrimination threshold drops below the detection threshold (facilitation effect) and rises as base contrast increases (masking effect). This pattern of results is known as the dipper function. We present the results of two observers for psychophysical contrast discrimination experiments that used natural images, rather than sinusoidal stimuli.
Stimuli were two digitised, linearised black-and-white photographs of natural scenes, and two random dot patterns that were filtered to have the same second order statistics as natural images. Experiments (following local guidelines) used a 2-alternative-forced-choice procedure in which the observer had to indicate which interval showed the base stimulus combined with contrast increment, as opposed to the base alone. The base stimulus was the full sized photograph, measuring 6 deg X 6 deg. The increment was a small, central, Gaussian-weighted patch of the photograph. Simultaneous staircases adjusted increment contrast for 8 base contrast values; performance converged on the 75 % correct, threshold rate.
For all 4 stimulus images, and for both observers, the plot of threshold increment contrast against base contrast showed both facilitation at low contrasts and high-contrast masking, but the amount of facilitation was less than obtained when we used sinusoidal gratings as base and increment. Facilitation was greatest for the random dot images, with discrimination threshold falling 8.0 dB below detection threshold for one observer, 6.0 dB for the other. Of the photographs, a street scene produced more facilitation than a garden scene (7.0 vs. 2.5dB for one observer, 4.5 vs. 3.5 dB for the other).
These data were compared with the predictions of a simple-cell computer model, based on the model of Foley (1994) and first developed for contrast discrimination experiments using sinusoidal stimuli. Good fits are achieved for the photograph experiments, but the model consistently underestimates the amount of facilitation that occurs in the random dot experiments. The results of the psychophysical and computational study suggest that similar neuronal mechanisms underlie the processing of contrast differences in sinusoidal and natural stimuli, and that understanding the experiments using sinusoidal stimuli can indeed shed some light on the processing of natural images. However, the differences between the data for the photographs and random dot images suggest that the natural scene data cannot be explained by reference to second-order statistics alone.
This work was supported by the MRC
Where applicable, experiments conform with Society ethical requirements