Modelling Fly Motion Vision - Computational Neuroscience: A Comprehensive Approach - page 395

Figure 14.2
Minimal circuit diagram of a correlation detector. It consists of two subunits. In each
subunit, the retinal signals from two neighboring locations are multiplied with each
other (M), after one or both of them have been fed through a temporal filter with a
time constant . This operation is done twice in a mirror-symmetrical way in both

subunits. The output signals of both subunits are finally subtracted.

τ

14.2 Mechanisms of local motion detection: the
correlation detector

The process of local motion detection has been successfully described by the so-
called correlation-type of motion detector or, in brief, correlation detector. This
model has been first proposed on the basis of experimental studies on optomotor
behavior of insects [49, 72, 73, 74, 75]. In subsequent studies, the correlation detec-
tor has also been applied to explain motion detection in different vertebrate species
including man (e.g., [34, 35, 87, 88, 89, 90, 96], for review, see [6, 8]).
In its most parsimonious form, such a correlation detector consists of two mirror-
symmetrical subunits ( Figure 14.2, left). In each subunit, the signals derived from
two neighboring inputs are multiplied with each other after one of them has been
shifted in time with respect to the other by a delay line or a temporal low-pass filter.
The final detector response is given by the difference of the output signals of both
subunits. The combination of a temporal delay and a multiplication is the reason why
this type of detector measures the degree of coincidence of the signals in its input
channels or, in other words, performs on average a spatio-temporal cross-correlation.
The basic operations of the correlation detector are summarized on the right side of
Figure 14.2. Here it is assumed, for simplicity, that the brightness distribution of the
retinal image is not filtered spatially or temporally but directly feeds the movement