Perception of transparency in stationary and moving images.

If two neutral density (ND) filters overlap partially, the luminance of the overlapping region is given by a multiplicative relationship (e.g., if the filters have 50% transmittance then on a 100 cd m-2 background, the luminance of each filter would be 50 cd m-2 but the luminance of the overlapping region would be 50% of 50 cd m-2, i.e. 25 cd m-2). Does the visual system respect this relationship? Two grey rectangles were overlapped to form a 'cross' and the luminance of the intersection was randomly varied. Naive subjects' ratings of transparency showed a surprising consistency with physics. A similar relationship is seen if two moving square-wave gratings are superimposed. When the intersection luminances were close to the multiplicative (25 cd m-2) case, component motion was seen, and values less than 25 cd m-2 and higher than 50 cd m-2 caused a decline in perceived component motion. Two interpretations are offered. (a) The visual system has access to 'tacit knowledge' of transparency and shadows. (b) If you assume a log signal compression in the retina, then by multiplying the luminances you would null the Fourier energy from the 'blobs' (i.e. regions of overlap between the two gratings). This would prevent 'blob tracking' and lead to the perception of coherent motion. Since most researchers in the field had simply added the gratings linearly they would have inadvertently introduced extra Fourier energy in the blobs and this might account for all previous instances of coherent motion observed in such displays. Whatever the ultimate interpretation, the present results provide the first clear evidence that even the processing of certain primitive visual dimensions--such as motion--can be powerfully constrained by the perception of transparency (Ramachandran V. S. (1990) in: AI and the Eye, Wiley, Chichester; Stoner, G., Albright, T. and Ramachandran, V. S. (1990) Nature 344, 153-155). This is well in line with other recent results demonstrating that transparency can also constrain the processing of stereopsis (Nakayama, K., Shimojo, S. and Ramachandran, V. S. (1991) Perception 19, 497-513) and perceptual grouping (Ramachandran, V. S. (1990) in: AI and the Eye, Wiley, Chichester).