Visual information is first processed in retinal ganglion cell (RGC), transmitted to lateral geniculate nucleus (LGN) of the brain and then to primary visual cortex (V1) in the cerebrum. These visual cells have limited fields of view, or windows, called receptive field. Neurons in RGC and LGN have a coaxial-shaped receptive field structure formed by the excitatory region and the inhibitory region. The on-centre simple cell receptive fields are elongated with an excitatory central oval, and an inhibitory surrounding region. Both receptive fields have antagonism between the excitatory and inhibitory regions. The RGC receptive field structure can be represented by DOG (difference-of-Gaussians) function, which is the difference between two Gaussian functions. The classical receptive field of simple cells can be approximated well with Gabor function.
In the cell membrane of Halobacterium Salinarum, photosensitive protein that precedes photosynthesis exists. This protein is called bacteriorhodopsin (bR), because it resembles the visual pigment rhodopsin of animals. BR shows differential response with a polarization reversal and we interpret this response as an equivalent for excitatory/inhibitory response of receptive filed. Employing this particular feature of bR we propose two types of optical filters, DOG and Gabor filters, and aim to apply for image processing.
We have fabricated receptive fields consisting of two oppositely coated bR films on the front and rear ITO plates. These films relate to the excitatory and inhibitory regions, respectively. We demonstrated temporal and spatial frequency characteristics of vision using such bR-based visual receptive fields. Our DOG filter, which mimics on-centre RGC receptive fields, had the function of a Laplacian filter and acted as an edge detector. The X-type receptive field responses obtained by the filter, for a variety of stimuli, were compared with available electrophysiological recoding. An on-centre Gabor filter strongly responded to a grating with particular spatial-frequency and orientation.
The goal of this research is the development of artificial DOG and Gabor optical filters for improving the analog image processing performance. Notably, the low cost and simplicity of fabricating single element bR-based filters, and no requirement for external connections are two of the major advantages over conventional silicon semiconductor technology.