| | ![]() | |||||||||
1996 International Topical Meeting on Optical Computing, Sendai, Japan, April 21-26.
2
ion-implanted MQW structure is low ~ 0.1mW/cm2 [4] and the
gratings for the high spatial frequency components in the input
image were saturated. In this case the carrier index gratings for
all the spatial frequency components have the same strength
which is equivalent to a thresholding operation on the power
spectrum. Additionally some hard clipping in the index variation
may be expected so that the device operates similar to the
thresholded and binarized JTC [5]. Figure 3 shows the
correlation result for two identical circles of light in the input
image, the experimental result with the MQW device is shown
above the simulation result for the thresholded and binarized
JTC.
We used a video-rate frame grabber card to capture the
image from a CCD camera and combined this with a previously
stored image using a scan converter. This was input to the SLM
allowing the correlation output to be recorded as the input was
changed in real time against a fixed reference image. Results
will be presented showing that the system showed high
discrimination between different face inputs. Figure 4 shows a
result of changing facial expression in the input. The correlation
output is influenced most strongly by the high spatial frequency
components in the input, i.e. its edges. A change in
facial expression still contains much of the original
edge information and a large correlation peak
results. Video input to the correlator allows
changes in the input to be tested simply, such as the
tolerance of the correlation signal to a few degrees
of rotation. We will report on the auto-correlation
and cross-correlation signals between different
individuals [6].
The photorefractive MQW device has
many advantages for a pattern matching system
based on the JTC. It offers a low power and
compact laser diode based system with diffraction
limited resolution and high correlation rate. It has
potential for high frame rate throughput for
applications such as searching large data bases.
Acknowledgements. The semiconductor MQW device was grown by MBE, and the thin film layers deposited, at the Institute for Microstructural Sciences of NRC.
References.
[1]. A. Partovi, et. al., Appl. Phys. Lett. 62, 464 (1993).
[2]. A. Partovi, A. M. Glass, T. H. Chiu, D. T. H. Liu, Opt. Lett. 18, 906 (1993).
[3]. S. A. Boothroyd et. al. paper WJJ2, W. S. Rabinovich et. al. paper WX6, OSA annual meeting, Sept
10-15, Portland, Oregon 1995.
[4]. A. M. Glass et. al., Opt. Lett. 15, 264 (1990).
[5]. B. Javidi, C. J. Kuo, Appl. Opts. 27, 663 (1988).
[6]. Video clips of the real-time face input tests will be available on the internet at http://alpha.ps.iit.nrc.ca.
Figure 3. Auto-correlation peaks for input of two identical circles of light. See text for details.
Figure 4. Auto- and cross-correlation peaks from the face recognition system.