The name is absent

121

intensity. Because natural scenes rarely show more than 50% contrast [105], there is little
more than a 0.3 log unit difference in light intensity from the darkest to brightest part of a
scene. There are likely some exceptions to this finding, but it means that the photon noise
driving the photocurrent in rod photoreceptors would be nearly uniform over the retina for
a given mean light intensity of a visual scene. As a result, the simple analysis performed in
section 5.6, is also applicable to extrinsic photon noise as well as intrinsic noise.

In the future, it would like to extend the analysis of noise in the rod network presented
here to demonstrate how the signal-to-noise ratio varies as a function of not only spatial
frequencies, but also light intensity. It may also be useful to consider the exceptional cases
where large contrast differences in a single image cause differences in background noise over
different parts of the retina.

Another logical next logical step would be to extend the analysis of filtering in the rod
network to investigate how the noise in rods is transformed and added to by higher order
neurons in the retina. Specifically, it would be interesting to investigate the tradeoff of spa-
tial averaging in the rod network vs. spatial averaging in the rod bipolar cells. Ideally if the
rod-bipolar cell synapse were linear, there would be no difference between the two. How-
ever, given that the synapse has been shown to be nonlinear, there would be a difference in
the averaging done before and after the synapse. Tessier-Lavigne and Attwell analyze this
paradox using an ideal coupled network [101]. Their explanation is that at at the initial rise
of the sigmoidal response curve characteristic of the rod-bipolar cell synapse, the gain for
small signals is lower than that for incrementally larger signals. This means that signals rid-

<<   127   128   129   130   131   132   133   >>

More intriguing information