Statistical Bias Predicts Many Illusions

There is a principle underlying visual computations that previously has not been recognized. This principle is about the effects of uncertainty. Many visual computations are estimation processes. Because of noise, and there are many sources of noise in the formation and processing of images, systematic errors occur in the estimation. In statistical terms we say the estimation is biased. To avoid the bias would require accurate estimation of the noise parameters, but this because of the large number of unknown parameters (the geometry and photometry of the changing scene) in general is not possible. Visual computations, which are estimation processes include the low level processes of feature extraction and the middle level processes of visual recovery. We hypothesize that the bias in the estimation of image features, that is points, lines, and image movement, is the main cause for most geometrical illusions as well as illusory motion patterns. Because of bias the location of image features is estimated erroneously and the appearance of patterns is altered. It is shown that many geometrical optical illusion patterns are such that the bias is highly pronounced. We analyzed the bias in visual shape recovery processes and found that it is consistent with what is empirically known about the estimation of shape. It has been observed from computational as well as psychophysical experiments, that for many configurations there is a tendency to underestimate the slant. The bias predicts this underestimation of slant. To demonstrate the power of the model we created illusory displays giving rise to erroneous shape estimation.