BRIGHTNESS ESTIMATION IN A NEURAL NETWORK MODEL WITH PRESYNAPTIC INHIBITION
AbstractRecent psychophysical and neurophysiological investigations showed that visual systemencodes luminance and use it to estimate illumination and surface brightness. We proposed anovel neural model for luminance coding based on recurrent inhibition, from the retinalganglion cells to the axons of the bipolar cells, which modulates the amount of sensory inputthat ganglion cells receive (Sagdullaev et al., 2006). Extended version of the model, where theamount of presynaptic inhibition is made proportional to the maximum luminance in thevisual scene, implements gain control mechanism which adjusts the raw luminance into ameasure of brightness of surface. Computer simulations showed that the model scalesbrightness estimates consistent with the highest-luminance-as-white anchoring rule (Gilchistet al., 2004). Simulations also showed that the model is able to act as a change detectorwhen the presynaptic inhibition temporally lags behind the excitatory input to the ganglioncell.