A Neurobiologically and psychophysically consistent model of the vertebrate retina
W H A Beaudot, J Hérault (Laboratoire de Traitement d'Images et Reconnaissance de Formes, Institut National Polytechnique de Grenoble, 46 avenue Félix Viallet, 38031 Grenoble Cedex, France;
e-mail: ; WWW: http://wbeaudot.kybervision.net)
The processing performed by the vertebrate retina begins to be well understood although its influence on subsequent stages still remains unclear. In order to clarify the retinal function in the visual perception, we propose one approach which attempts to infer functional properties from the neural architecture of the retina. Our retinal model is simple enough to be analysed with the classical signal processing tools, and realistic enough to have a functional relevance. This model consists of two functional layers, the Outer Plexiform Layer (OPL) and the Inner Plexiform Layer (IPL). Each of them is made up of several neural layers. Each neural layer is modelled by a resistive grid with different spatial characteristics reflecting the coupling between neighbouring neurons of the same layer. Each neuron is finally modelled by a leaky integrator.
The antagonistic interaction between the photoreceptors and horizontal cells layers (in OPL) gives rise to an inseparable spatiotemporal bandpass filter. So, in the signal domain (e.g., for image sequences) this filter realises a contrast and motion enhancement. This output of the OPL filter conveys the X-type retinal response into the IPL through the bipolar cells layer. The processing performed by the amacrine and ganglion cells layers (in IPL) was theoretically deduced and correlated with the known data concerning its function in the motion analysis (detection of motion and directional selectivity). The specific IPL processing provides the Y-type and W(DS)-type retinal responses. To complete this retinal model, chromatic and adaptive features have also been added at the level of the photoreceptor.
Moreover, the model has also led to a realistic simulation tool for early visual processing including space, time and colour data. Since this structural model has shown properties similar to those provided by neurophysiology and psychophysics, it allows to expect a reduction in the gap between neurobiological and psychophysical data. Therefore, this model (and its simulation tool) can provide a realistic output signal of the retina, and it can be used as the input signal of farther structures.
© 1997 CSEM S.A.