Glass patterns: some contrast effects re-evaluated.

The relative contrast of features is known to be important in determining if they can be grouped. Two manipulations of feature contrast have previously been used to criticise models of visual grouping based on spatial filtering: high-pass filtering and reversal of contrast polarity. The effects of these manipulations are considered in the context of the perception of Glass patterns. It is shown that high-pass filtering elements, whilst destroying structure in the output of low-pass filters, do not significantly disrupt the output of locally band-pass filters. The finding that subjects can perceive structure in Glass patterns composed of high-pass features therefore offers no evidence against such spatial filtering mechanisms. Band-pass filtering models are shown to explain the rotation of perceived structure in Glass patterns composed of opposite contrast features. However, structure is correctly perceived in patterns composed of two 'interleaved' opposite contrast patterns, which is problematic for oriented filtering mechanisms. Two possible explanations are considered: nonlinear contrast transduction prior to filtering, and integration of local orientation estimates from first-order and second-order mechanisms.