engleski

New Mechanism for Luminance Channel in Network Model of Brightness Perception

Neumann (1996) proposed an extension of Grossberg-Todorović model of brightness perception that includes separate luminance channel. Activity distribution in luminance channel corresponds to the nonlineary compressed input luminance distribution. This is achieved by additive pooling of the ON and OFF retinal ganglion cells responses. However, such scheme could not support adequate luminance difference processing if parameter setup allows ON and OFF cells to extract ratios between surfaces. That is, enlarging the luminance difference between surfaces by enhancing the illumination should enlarge difference in activity distribution in luminance channel while activity in contrast channel remains constant. Importance of luminance difference processing is highlighted by the fact that at the intersection of illumination and reflectance edge ratio remains invariant while luminance difference change (Gilchrist, 1988). Here, it is proposed that luminance channel could incorporate cells with similar receptive field structure like retinal cells with preprocessing by power law transformation in excitatory and inhibitory synapses. Power law describes relation between amplitude of presynaptic and postsynaptic potential. In order to achieve desired functional properties we need to impose additional assumtion of greater nonlinearity in excitatory synapses. Computer simulations show that network adequatly processes luminance differences. That is, larger luminance differences in input correspond to larger differences in the activity distribution. Besides, network responses are stronger at the edges than at the uniform regions. In this way, network transmits separate contrast information which is correlated with luminance, rather than low-pass filtering isomorphic to luminance. This is consistent with Whittle's (1994) conclusion that cells in visual patway may trasmit different contrast signals for the same stimulus. Asymmetrical power law transformations could also be used for modeling simple cells receptive fields. Recent neurophysiological investigations (Rossi et al., 1996) reveals that large proportion of cells which has been traditionally considered as a orientation selective actually shows multiplexing of orientation and brightness. Moreover, they exhibit sensitivity to luminance variations outside their classical receptive fields. Such receptive field properties are obtained by asymmetrical preprocessing in excitatory and inhibitory synapses before convolution with isotropic kernels. Further research will explore possible interactions between orientation selective cells in luminance and contrast channels and their functional significance.

brightness perception; luminance; neural networks; power law

nije evidentirano