BERKELEY, Calif. -- A little-known device called an optical correlator is helping to drive the conversion of all-optical processing systems into powerful processing products for such lucrative commercial applications as fingerprint recognition and medical imaging. The miniaturization of optical components and their hybridization with digital electronics promise a new wave of systems that could far exceed the proc essing power of today's electronics.

Computer theorists and high-speed-coprocessor engineers have been tantalized for decades by the high-speed potential of optical computing systems, but historically the technology has failed to yield products that could compete with electronic systems in cost and ease of use. Now, broad-based advances are yielding compact optical processors that can tackle such computationally thorny tasks as image recognition. Optical computing systems that once would have occupied the surface of a tabletop are now being fit into form factors suitable for desktop computers. And a number of companies are eyeing the commercial potential of dedicated high-speed optical processors.

The push toward optical computing is scaling down optical correlators, a mature technology developed by the military, to address fingerprint-recognition and other pattern-matching applications. Optical correlators execute the Fourier transform with parallel optical operations to perform comparison operations instantly on entire images. In the 1980s, such systems were huge and inefficient, occupying whole optical tables in labs. Today an optical correlator can be placed on a card that fits inside a personal computer.

Developers are also exploiting digital electronics to make their products competitive and easy to integrate into existing systems. The result is hybrid pattern recognizers, processors and memories that represent the best of both optical and electronic worlds. If successful in the marketplace, the compact correlators will have proven that optical computing has a future.

The mood is optimistic--so much so that even Stuart Schoenmann of OCA Applied Optics (Garden Grove, Calif.), which has just exited the optical-correlator business after a decade, is confident of the market's potential. An assessment to the contrary by parent company Corning, which took OCA over last April, has not dampened Schoenmann's enthusiasm. "We continue to get interest from different areas," he said. "We're hopeful for the industry and for the technology."

Further integration of tiny lasers and other optical components, Schoenmann said, will produce a "huge jump in technology and capabilities. We're excited about that."

SLM availability
One impediment to the commercialization of optical correlators has been that the devices' main components are spatial light modulators. SLMs are crucial because they allow the input images and filters to be encoded onto beams of light. But until recently, small, cheap, electrically addressable SLMs with high pixel counts were simply not available. Today they are--as are the tiny diode lasers and low-cost single-chip cameras that most correlator systems also require.

Some correlator designs also need an optically addressed SLM (OASLM) to record the interference pattern between two transforms. Those are still somewhat rare and expensive so the companies that have succeeded in building the processors have had to develop their own OASLMs.

For instance, QuantaImage (Bu rlington, Mass.) was formed because its parent company, CoreTek Inc., had developed quantum-well-based optically addressed modulators. Those had the advantage of being fast, with high resolution.

Lucrative application
QuantaImage's system was designed to address one of the security industry's most lucrative applications to date: fingerprint recognition. Called the FPID-256, the machine can match images at a rate of 1,000 frames/second. In addition to speed, QuantaImage's access to good OASLMs has afforded flexibility in image input. Rather than use an electronic camera to capture data, for instance, light from the outside world is focused onto the modulator itself. The image is then picked up by the correlator's laser beam and fed directly into the system. That eliminates two steps: image acquisition and image transmission.

At Hamamatsu Photonics K.K. (Hamakita City, Japan), the OASLM is based on ferroelectric liquid crystals. Like the CoreTek device, the Parallel-Aligned Liquid Crystal Spa tial Light Modulator (PAL-SLM) is both high-resolution and fast. In fact, high resolution is generally easier to achieve with OASLMs because they require no driving electronics or communications wires and thus do not have to be structured as pixel arrays.

Hamamatsu was an early player in fingerprint recognition; its systems have been sold commercially and have been installed at various banks and other security-conscious companies. As well as using a fairly standard setup, in which an input from the real world is compared with one presented to an SLM as a video signal, the company has been involved in a collaboration with Holoplex (Pasadena, Calif.) to store known fingerprint images holographically. The system saves time because it presents a full image immediately (instead of loading it line by line into the SLM), and it removes some constraints on the size of the fingerprint databases that can be searched.

Another company with a fingerprint-recognition system that is up and running is Mytec Technolog ies Inc. (Toronto). Called the True Recognition System, it uses a standard optical setup known as a Vander Lugt correlator. A person who wishes to be identified places a finger on a prism, which is illuminated from one side by a laser beam. Where the skin touches the glass, light is coupled out through refractive-index changes created by the skin's moisture in contact with the glass surface. The rest of the light is totally internally reflected. The laser beam thus picks up an image of the fingerprint, which can be transformed and compared with various prerecorded filters.

Optically, the system is relatively simple and uses off-the-shelf components. The digital signal processing system that is used to analyze the correlator output is more complex, using a field-programmable gate array to provide a PC with various statistics about the incoming correlation peaks. The FPGA can do that in real time (60 frames/s) because it can rewire itself to create a dedicated circuit for each new calculation. That lets the correlator operate at full speed. In turn, the PC can quickly determine the user's identity.

The fact that correlator research has traditionally been government-funded partly explains the glut of fingerprint recognizers on the market. In the post-Cold War world, identifying terrorists and criminals has become more crucial than targeting tanks and planes. But many other applications are taking shape.

Defense-conversion Rx
For instance, when forced to look outside the military for expanding market areas, Lockheed Martin (Denver) turned to medicine. Its designers have teamed with researchers from Rose Health Enterprises to develop an optical processor to identify cancerous breast lesions. Again, the performance of the basic optics is enhanced using digital electronics, this time in the form of a neural network to extract meaning from the correlation peaks.

The current setup already performs 10 percent to 30 percent better than the average radiology-based method, the company claims. It is also said to operate 10 times faster than existing systems while incurring only half the expense.

Lockheed Martin has also entered a partnership to inspect the placement and soldering of electronic parts on assembly lines, according to Scott Lindell at the company. "The system is designed to be able to inspect an area of approximately 18 inches x 18 inches with about 1,000 parts and 8,000 solder joints in less than 20 seconds," Lindell said. "The system will determine whether all the correct parts are present on the board and properly position them within manufacturing tolerances. It will inspect at much finer resolution to provide statistical-process-control information to support planned assembly-line maintenance and to predict failure before it happens."

The system may be used in conjunction with a neural network and is about 60 percent developed.

Another company that has used the military as a training ground for optical systems is Litton Data Systems (Agoura Hills, Calif.), developer of the M iniature Ruggedized Optical Correlator. The MROC has the advantage of being completely solid: The path between laser, lenses and SLMs is composed not of air, but of a sturdy transparent material that "binds" the device together.

In addition to their work on applications, researchers at Litton are developing tools to let potential customers test various optical-correlator configurations, algorithms and filters themselves.

David Carrott, recognition-systems program manager at Litton, said a crucial step in the transition from digital signal processing is "getting people to start thinking about optical signal processing." That, Carrott said, is where Litton's simulation package, the Optical Recognition System Simulator (ORSS), comes into play.

The object is to take the expertise that's developed in digital signal processing and bring it over into the world of optical signal processing--involving algorithms, filters and the techniques of the hardware involved around optical correlators," Carrott sai d. To push the work forward, Litton Data Systems is putting together a course that will explain how the ORSS works. The program is designed to seed the market for optical correlators.

More than recognition
Though optical correlators excel at fast pattern-matching, the optical processors can be much more than just a recognition system. Joint transform correlators, for example, can be particularly useful for tracking elements that are moving in time. By correlating successive video frames, for instance, the distance between two otherwise identical images can be found.

Hamamatsu has exploited that fact to perform displacement and velocity measurements on speckle patterns to show, for instance, how various kinds of particles flow.

Physical Optics Corp. (Torrance, Calif.) has applied the technique to eye-tracking. Cameras are trained on the subject's eyes, and an optical novelty filter digitally calculates "difference" images between subsequent frames. (Only pixels that have changed appear in the filtered image.)

One difference image is loaded into the input spatial light modulator; its predecessor is used as a reference. The two interfere to produce a joint power spectrum (Fourier transform), which is then analyzed digitally using a DSP module.

The eye tracker has the advantage of being fast, high-resolution and relatively passive (the light used poses no special problems for the eye).

For the future, some companies are looking into even more advanced and complex applications. Yoshiji Suzuki, director of Hamamatsu's Central Research Laboratory, looks at optical correlators as a step on the road to practical optical neural computers, i.e., adaptive systems containing optical correlators that search optical memories and have the capacity to learn.

Sunny Bains is a freelance science writer/technical editor based in Berkeley, Calif.

  Go To Previous Story

Go To Next Story  

Back to This Week's News