Reichardt Detector, Exponential Temporal Filter
This is a Reichardt detector (RD) model for direction selectivity
(Reichardt, 1957; 1961) with Poisson spike generation.
It consists of a set of twenty RD subunits aligned parallel and
sequentially along the axis of preferred motion. Each subunit is an
opponent mechanism that senses the visual input via a pair of
point-like spatial detectors and uses multiplication to combine
signals that are processed by either a low-pass filter or a high-pass
filter, both implemented with one-sided decaying exponential
functions. This form of RD model is similar to that used by Borst et
- Borst A, Flanagin VL, Sompolinsky H (2005) Adaptation
without parameter change: dynamic gain control in motion detection.
- Reichardt W (1957) Autokorrelationsauswertung als
Funktionsprinzip des Zentralnervensystems. Zeitschrift für
Naturforschung, Teil B. 12:447--457.
- Reichardt W (1961) Autocorrelation, a principle for the
evaluation of sensory information by the central nervous system. In:
Sensory Communication, Rosenblith WA, ed. pp 303--317. MIT Press,
Each RD subunit processes the signals from two neighboring
stimulus pixels (separation dx
= 1 pix). The time varying
stimulus is convolved with a low-pass (LP) filter that is a one-sided
decaying exponential and a high-pass (HP) filter that is a
delta-function minus a decaying exponential (see equations, upper
right). The LP filter delays the signals in both the left and right
branches before it is multiplied to form the preferred, rp
responses, respectively. The difference of
these responses is the final (opponent) response, ropp
(B) The overall model output is the sum of the ropp
signals from multiple subunits aligned along the x-axis, sharing
neighboring pixels, as shown.