CMOS
dynamic linking networks for real-time human face tracking
- Spencer, R.G.
Analog & Mixed-Signal Center, Texas A&M; Univ., College
Station, TX, USA
This paper appears in: Circuits and Systems, 2000.
Proceedings. ISCAS 2000 Geneva. The 2000 IEEE International Symposium on
On page(s): 427 - 430 vol.1
28-31 May 2000
Geneva, Switzerland
2000
Volume: 1
ISBN: 0-7803-5482-6
IEEE Catalog Number: 00CH36353
Number of Pages: 5 vol.(viii+813+768+769+768+760)
References Cited: 8
INSPEC Accession Number: 6709843
Abstract:
Dynamic linking networks are highly effective in locating human
faces, as they are invariant to changes in position, orientation,
scale, deformation, and partial lighting. Additionally, these networks
are well suited for analog VLSI using very simple CMOS circuits such as
programmable ring oscillators, strongly coupled diffusion lattices, and
AC-coupled Hebbian synapses. Dynamic linking networks are orders of
magnitude faster than digital computers, thus they provide a fast,
efficient, and robust means of tracking faces in real-time.
Transistor-level simulation results are presented for the .5 /spl mu/m
TSMC CMOS technology with 3-volt supply.
Index Terms:
real-time systems; face recognition; feature extraction; VLSI;
analog processing circuits; CMOS analog integrated circuits; Hebbian
learning; neural chips; CMOS dynamic linking networks; real-time human
face tracking; position; orientation; scale; deformation; partial
lighting; analog VLSI; programmable ring oscillators; strongly coupled
diffusion lattices; AC-coupled Hebbian synapses; transistor-level
simulation results; TSMC CMOS technology; 0.5 micron; 3 V