Special Report: Optical Patterns
12.3.2 Classification of optical pattern recognition systems Optical correlators have been around since the early 1950's. Although most of the current optical correlators are based on the coherent optical processors (either the VanderLugt or the joint transform correlator), there are many other types of correlators, as is shown in Fig. 12.8. Below, a brief description of each of the systems is provided. More detailed analyses can be found in various Ref's [46-48].
Fig. 12.8. Classification of optical correlators. TDI, time delay and integration.
(1) Kretzmer-type correlator [49]: This is probably the oldest and most straightforward correlator developed in the early 1950's. In the system, summation is achieved spatially by an optical lens and is detected by a single detector. Also, a shift operation is achieved by one (or both) of the inputs is mechanically moved along both the x and the y directions. Such a movement can be achieved by an acousto-optic beam deflector in which the signal flows through the one-dimensioal medium. Unfortunately, there are no fast 2-D image shifters, currently.
(2) Shadow casting [50-52]: In this system architecture, both shift-and-add operations are done in an all-optical fashion simultaneously. The problems with such a system are diffraction effects and no room for the spatial filtering operation. Microlasers, even with their natural 2-D nature, can find limited application for this type of correlator because of the diffraction effect that is due to the high coherence of the light from microlasers.
(3) Electro-optic correlator with a time-delay-and-integration-mode CCD [53,54]: In this system, an input signal is temporal and is emitted from a point source such as an LED or a laser diode. The reference signal is recorded in an SLM that is in contact with (or imaged to) a time-delay-and-integration-mode CCD sensor. Both the input signal and the CCD are synchronized and systolically perform the shift-and-add operation to calculate the correlation.
(4) Electro-optic correlator with a normal CCD: This system is equivalent to the system in (3) except that the time-delay-and-integration-mode CCD is replaced by a moving reference signal and a normal CCD.
(5) VanderLugt correlator [55]: This architecture has probably been the most commonly used architecture for optical pattern recognition. Pattern recognition is achieved in two steps: (1) holographic filter fabrication and (2) processing. In the first holographic filter fabrication step, a matched filter for a given reference image is holographically fabricated with the Fourier transform holographic recording technique. In the processing step, if an input matches the reference image, the wave-front distortion generated by an input pattern is canceled out and a plane wave is generated. The plane wave is then focused to a point at the focal plane of the second lens. The system has been successful because of the high signal-to-noise ratio (SNR) that can be achieved with additional spatial filtering that can be easily achieved in the system. Also, once a holographic filter is fabricated, it is fully nonblocking without requiring a photon-to-electron conversion process that can lead to speed bottleneck. However, the system requires an input from high-optical-quality recording materials and a holographic filter to preserve shift-invariant recognition. The filter has to be positioned exactly in the original position in which it was recorded. Also, the holographic filter must be positioned exactly in the focal plane of the Fourier transform lens. Recent advances in high-quality SLM's and in situ holographic recording materials allow a convenient implementation of such an optical correlator with high performance.
Previous Next Table
of Content for report: Optical Patterns Home
