A computational analysis of separating motion signals in transparent random dot kinematograms

When multiple motion directions are presented simultaneously within the same region of the visual field human observers see motion transparency. This perceptual phenomenon requires from the visual system to separate different motion signal distributions, which are characterised by distinct means that correspond to the different dot directions and variances that are determined by the signal and processing noise. Averaging of local motion signals can be employed to reduce noise components, but such pooling could at the same time lead to the averaging of different directional signal components, arising from spatially adjacent dots moving in different directions, which would reduce the visibility of transparent directions. To study the theoretical limitations of encoding transparent motion by a biologically plausible motion detector network, the distributions of motion directions signalled by a motion detector model (2DMD) were analysed here for Random Dot Kinematograms (RDKs). In sparse dot RDKs with two randomly interleaved motion directions, the angular separation that still allows us to separate two directions is limited by the internal noise in the system. Under the present conditions direction differences down to 30 deg could be separated. Correspondingly, in a transparent motion stimulus containing multiple motion directions, more than eight directions could be separated. When this computational analysis is compared to some published psychophysical data, it appears that the experimental results do not reach the predicted limits. Whereas the computer simulations demonstrate that even an unsophisticated motion detector network would be appropriate to represent a considerable number of motion directions simultaneously within the same region, human observers usually are restricted to seeing not more than two or three directions under comparable conditions. This raises the question why human observers do not make full use of information that could be easily extracted from the representation of motion signals at the early stages of the visual system