🚀 go-pugleaf

           Department of Computer Science, University of Toronto
            (LP = D. L. Pratt Building, 6 King's College Road)

       -------------------------------------------------------------

                                  VISION
               LP266, at 2:00 p.m., Tuesday 01 December 1992

                            Berthold K.P. Horn
                   Massachusetts Institute of Technology

                     "Analog VLSI for Machine Vision"


Analog VLSI has the potential for fast, small, low power, cheap front end
vision processing that does not require a great deal of support circuitry.
But it is hard to compete head on with digital VLSI, which has a much longer
development history.  So merely implementing existing algorithms --- which 
were developed in the context of sequential digital machines --- is not a 
good idea.  One needs to find ways to exploit the medium.  What are analog 
networks good at computing?  Several devices have been built that exploit 
properties of resistive sheets, theorems for converting area integrals into 
contour integrals, properties of Gaussian integers, switched capacitor 
networks and so on.  Possibly the most promising area is in real-time 
feed-back involving motion vision, where the long latency of competing 
pipelined digital systems is deadly.