A biologically inspired edge detection technique is presented, based on rank order coding and a spike train representation of data. Pixel intensity values, encoded as a succession of pulses, form the input to an array of edge detecting neurons. By comparing the order in which the input spikes are received to a stored edge profile, the firing frequency of the edge detector can be controlled. The edge detection system is shown to perform as designed, with the output activity of each edge detector corresponding closely to the clarity and orientation of the edge segments presented. One particularly interesting feature of the system is its ability to produce a meaningful output based upon very few input spikes and then improve this initial estimate as more information is received.
展开▼
