The WAVE Report on Digital Media
3D --- Media Creation --- Shared Space
---Published by 4th Wave, Inc.---
Issue #0335------------------11/03/03

 

The WAVE Report is Searchable on

http://www.3dlinks.com
http://www.wave-report.com

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

0335.1 Hot Topics

0335.2 Story of the Issue

0335.3 Semiconductors

0335.1 Hot Topics

***Caligari Announces Winners of ''Big Break'' Animation Contest
(October 28)

Caligari, a producer of 3D modeling and animation software, today released the winners of its Big Break 2003 animation contest. The contest was designed to showcase the talents of artists and animators using Caligari's trueSpace 3D modeling and animation software package, and illustrate what's possible using a desktop PC and a healthy dose of talent and imagination.

Seattle-based Spencer Britton, mechanical engineer by day and animator by night, won the Big Break's "Art Form" category with his animated short "Bombastic," a look into the adventures of the life-forms on an alien planet. Saul Greenberg, a London-based painter of set designs and other industrial projects, won the "Story Form" category for his animated short, "UMMM," a study of animators' block.

Greenberg created "UMMM" primarily on his laptop as he traveled between assignments rendering completed images on "the closest desktop" as time allowed. Britton also produced "Bombastic" during nights and weekends, using plug-ins including Motion Studio and Primal Particles FX from Primitive Itch and Space-Time Morph from Mente Magica for fine-tuning. Britton says his inspiration for animations is to make people laugh using unusual characters and machines with familiar qualities.

http://www.caligari.com/Gallery/press_animations.html


***NTT, Motorola Team Demo M-MPLS Broadcast Technology
(October 27)

NTT and Motorola, Inc. today announced a public demonstration of emerging M-MPLS (Multicast Multiprotocol Label Switching) technology. M-MPLS is a broadband transport mechanism that enables new network services requiring broadcast and quality of service features that are not possible with current multicasting technology. The two companies are developing a number of new service applications, the first of which are video distribution over Digital Subscriber Line and Fiber to the Home (FTTH), which will deliver rich media services to consumers.

NTT Network Service Systems Laboratory -- a research arm of the NTT Group -- and Motorola Computer Group, a supplier of industry standards-based embedded systems and value-added services, have collaborated to define the M- MPLS specification and move it into the IETF (Internet Engineering Task Force) standardization process, where it continues to gain support. At MPLS 2003 International Conference in Washington, D.C., NTT and Motorola will demonstrate a network of PC routers implementing the draft specification and a simulation of the control plane. Motorola intends to offer functionality such as M-MPLS in a range of platforms based on the AdvancedTCA specification. Motorola will also offer this draft specification M-MPLS software through its NetPlane software product line.

http://www.motorola.com


***Interactive Intelligence Develops New Speech Recognition Capabilities
(October 30)

Interactive Intelligence Inc., a developer of software for IP telephony, contact center automation and unified communications, has developed and filed patents for new speech recognition capabilities to appear in the next release of its product line.

Beta shipments of this latest release, version 2.3, have already begun and initial customer installations are expected before the end of the year.

The new software includes capabilities such as:

- An engine-agnostic interface that enables speech-enabled applications to be independent of a particular speech recognition engine
- Built-in support of the Scansoft (formerly SpeechWorks), Nuance and Aculab engines
- An N+1 architecture that enables speech recognition sessions to be handled by a number of servers for scalability and fault tolerance
- The ability to simultaneously deploy speech recognition engines from different vendors, and to use rules to determine which sessions are serviced by which engine.
- Support for the Speech Recognition Grammar Specification developed by the World Wide Web Consortium
- Speaker identification and verification

The first product to take advantage of the new speech recognition features will be version 2.3 of Interactive Intelligence's contact center automation software, Customer Interaction Center (CIC). Releases of the company's enterprise IP telephony software, Enterprise Interaction Center (EIC), and its unified communications software, Communité, will follow in the first half of 2004.

All three products will support speech recognition across a variety of telephony architectures, including traditional circuit-switched, voice over IP using session initiation protocol, and Cisco's AVVID platform. Interactive Intelligence also plans to add support for VoiceXML and Microsoft's SALT in the first half of 2004.

http://www.inin.com

0335.2 Story of the Issue

***Biometrics Consortium 2003
By James Sneeringer

September 22-24
Crystal City, VA

This was an event that was in many ways the reverse of our typical experience at technology conferences. Biometrics is an emerging technology industry, but unlike others we cover it is being primarily driven by the government, both in terms of the market, and to some extent in terms of the technology. The lead agency is NIST, the National Institute of Standards and Technology, which has been working on the science of biometrics for decades. In addition, the largest biometric identification operation in the country is the FBI's fingerprint program, which is headquartered at the Criminal Justice Identification Services Division, in Clarksburg, West Virginia. The Defense Department and national security agencies are some of the largest consumers of biometric technology. And, the State Department is tasked with enhancing U.S. entrance visas with biometric data, beginning in 2004.

Yet, the industry is already looking beyond the government to the enterprise, and to a lesser extent, the consumer. At the enterprise level, biometric identification is already being used to augment physical security--to control who gets in to the building or the room. Some businesses are also looking at biometric identification to enhance their data security--to control who can log into the network, and who has access to what data. Several companies on the exhibit floor were focused primarily on integration services for businesses.

In the consumer space, biometric identification is already being used for physical security by Disney, which uses a two-finger fingerprint identification system to confirm the identity of season pass holders at their theme parks. Several companies are offering USB flash memory dongles that protect their contents with built-in fingerprint-identification scanners. The Compaq iPaq PDA is now available with a built-in fingerprint reader to control access. Some states are considering adding biometric data to driver licenses and state ID cards. There is some talk of using biometric data as an authentication tool for online purchasing, or for operating system log-on. Some versions of Apple OS 9 included voice identification for log-on, but it has been removed from more recent versions.

Biometrics Overview

Peter Higgins of the Higgins-Hermansen Group, a biometrics consulting firm, started off the conference by presenting an extensive overview of biometrics, and an introduction to the different technologies, important issues, and terminology.

What is a biometric?

Biometrics are automated methods of recognizing a person based on a physiological or behavioral characteristic.

This can include any characteristic significantly unique, such as face, fingerprints, hand geometry, skin morphology, handwriting, iris, retina, vein, and voice. The three most commonly used are:

Fingerprint

This can be one to many fingers. Disney uses the first two fingers of the right hand for their season pass holders. The FBI enrolls all fingers and thumbs, and can match against any single or multiple of acquired data. Most consumer or enterprise solutions use the index finger of either hand. There is also a behavioral aspect to fingerprints, in that they can be collected either with a flat imprint, or rolled across the width of the whole finger pad.

Fingerprints can be gathered in a number of ways. The most well-known is the traditional law-enforcement fingerprint card, created on paper with black ink. This can then be scanned at high resolution into a database. Modern electronic sensors can be based on several technologies:

- Optical -- The sensor takes a monochrome image or images of the finger pad.
- Ultrasonic -- The sensor uses high-frequency sound waves to image the finger pad.
- Thermal imaging -- The sensor uses the thermal characteristics of the finger to image the finger pad.
- Capacitive -- The sensor is a semiconductor that uses an electromagnetic field to image the finger pad.

Fingerprint image analysis has typically been based on analyzing events in the fingerprint, known as "minutae"--areas where ridges meet, divide, end, begin, or are cut or broken. The type and location of the events are used to create the template that the system matches against. This system is what FBI recognition is based on. However, a new technology created by Bioscrypt uses an algorithm that analyzes the entire image of the print. This technology, known as correlation, won the most medals at the 2002 Fingerprint Verification Competition, for both speed and accuracy.

Fingerprints are the most mature biometric technology. They have been used by law enforcement for decades, and the FBI has over 40 million prints in their database. The best real-time implementation of fingerprint analysis is at Disney Orlando, where they use 2 fingers to identify season pass holders and process 2 million transactions per year. The response time for verification is now 11 seconds per attempt.

Iris

This uses an imaging device to take a picture of the eye, and analyze the patterns in the iris. The technology was first developed by a series of companies in the mid-1990s, that eventually merged to form the company Iridian. Almost all iris recognition technology in the marketplace today is based on Iridian's solution, which it licenses as software to hardware companies, who produce the actual imaging devices. The recognition is based on a combination of minutae and correlation. Several vendors and system integrators agreed that iris is the most accurate--it has the lowest error rates. However, it is difficult for users to self-help and is prone to failure-to-acquire errors.

Facial recognition

This uses a camera to capture an image of the person, and image analysis techniques to first isolate the face from the rest of the image, then reduce it to a template that can be compared to the enrollment data. The key to these systems is being able to find the eyes. Once the system can find the eyes in a picture, it can manipulate the image to rotate or transform it to the correct size for comparison. This technology developed in DARPA in the mid-1990s, and led to commercial products by 1997.

Identification by facial recognition is the only biometric that can be easily checked and confirmed by a person. This may make it ideal for systems that deal with the public, as error can be solved quickly by staff with minimal training--we are hard-wired to be able to recognize faces. However, facial recognition is one of the least-accurate biometric technologies. It is covered in more detail below.

To participate in a biometric identification program, a person first enrolls in the program, by providing the biometric data that will act as a basis for comparison. Typically the enrollment data is acquired under ideal circumstances, to provide the best possible baseline. From then on, whenever that person interfaces with the biometric system, a new set of data is acquired, and compared against the baseline, to produce a score. If the score is above a certain statistical threshold, the data is considered a match.

The key to the system is the statistical threshold for determining matches. If it is set too high, the system will produce errors known as false non-matches, or false rejects--a match is not indicated when it should be. If the threshold is set too low, the system will produce errors known as false matches, or false accepts--a match is indicated when it should not be. Another, more basic error is the failure to acquire--the system cannot gather sufficient data from the person to even attempt a match. The most basic error in a biometric system is the failure to enroll--the system cannot gather enough data to even form a baseline for that person. An obvious example would be a person missing a hand, who is asked to enroll in a hand-geometry biometric identification system.

Almost all the presenters and booth personnel we spoke with referred to biometrics exclusively in a managed security capacity--for site security, data access, or law enforcement. There was very little discussion of biometrics as it applies to consumer technology. The three applications most common applications for biometrics are:

- Watchlist--compare challenges against a database for a positive match
- Verify--compare challenges against a database and alleged identity to confirm identity
- Identifiy--compare challenges against a database to provide positive ID

Most biometric systems in use today for access control are based on verification. The person alleges a certain identity, using a pass code, key card, or id, and the biometric is used to confirm that identity. Because it is a straight one-to-one comparison, verification is the least taxing on the back-end systems. Watchlist is much more taxing, because the challenges are being run continuously against a list of possibilities, rather than just one. And identification is the most taxing, because each challenge is run against the entire database of enrollment data. In fact at this time, back-end database and computing systems are generally not robust enough to provide identification service on a wide scale.

Biometrics in general now face the problem of scaling. If used on a wide-scale basis, such as to screen airline passengers, even an accuracy rate of 95% will produce tens of thousands of travelers who fail to match and must be human-verified. In reality the accuracy rate is likely to be much lower, producing even more failures to identify. And at a certain point, system errors will over-ride recognition accuracy. The best estimate from testing is that small errors in the system such as data entry, poor capture or poor enrollment, can cause error rates of 1 in 1000. Therefore even an algorithm with perfect recognition could not exceed this error rate in practical use.

There is also the question of what databases are to be searched, and just as important, what would be done with the information the resulted? If an air passenger matches for a register sex offender and a warrant for parking tickets in Seattle, what burden does that place on the airport security forces? The refrain was repeated over and over--business systems, practices and rules may be more important to deployment success than the recognition technology itself. Most important are the error handling techniques. It reminded me a lot of handwriting recognition--since the machine/human interaction is inherently imperfect, the correction systems will define the ultimate value. At this point that is an issue that most vendors ignore completely.

Biometrics Testing and Evaluation

Much of the terminology of biometrics is not agreed upon yet, especially as it refers to testing and test results. This is especially true for the terminology associated with understanding error rates, such as the "false-accept rate."

Two presentations were solely devoted to terminology and testing problems. One presenter, Valerie Valencia, CEO of Authenti-Corp, stated that there has never been a small-sample test that scaled to predict large-scale results with any statistical significance. Her most incriminating statement was that some vendors deliberately abstain from independent testing, to maintain the illusion of accuracy through vague terminology. She made a point to state that this info came from the vendors themselves--with anonymity of course.

As it stands now, there are very real issues about the accuracy and scalability of the two primary biometric technologies--fingerprints and facial recognition. Fingerprints are by far the most mature of the biometric technologies, but they are slow and invasive to acquire, slow to process, and have accuracy issues as they scale to large populations. In addition, there is no agreement on data handshake standards, or even on the various techniques for enrolling--individual rolled, individual flat, together, 2-finger, etc.

Facial Recognition

Facial recognition has many more problems with accuracy. Many of the sessions referenced the recently completed FRVT (facial recognition vendor testing) 2003 by a coalition of organizations spearheaded by NIST. The testing was based on a database of 37,000 Mexican immigrant applicants. The best results were by Cognitec, using controlled indoor lighting, on the same day, facing head-on, and with a small database to match against. Accuracy was about 95%. As any factor changed, the results got significantly worse. A larger database had a negative effect, lowering the accuracy by up to 50% if all 37,000 were searched against. Outdoor lighting was an even worse factor, dropping all vendors to under 60% accuracy due to the harsh shadows, squinting, etc.

In public deployments, such as at airports or border crossings, the accuracy can be expected to be low based on the FRVT results. The enrollment image is likely to be of low quality (passport or drivers license photo), old (more than a day or two starts to affect error rates), the lighting will be different, as will hair style, facial hair, facial expressions, etc. In addition, women have a significantly lower accuracy rate than men, up to a 50% difference for young men and women. The differences are less, and facial recognition for both genders more accurate, for older subjects.

There was some positive news about facial recognition. First, morphable 3D modeling had a huge positive effect on faces that were not facing directly forward, sometimes doubling the accuracy over simply submitting the offset face directly to the algorithm. There was also news of a successful facial recognition deployment in Australia. It is run by the Australian government and used to verify identity for international flight crews for Quantas who are based out of Australia. The kiosk reads the passport via OCR on a reader, while capturing the face head-on under controlled lighting. Enrollment (initial data capture--the stored images that are tested against) was based on 5 images under the same lighting, taken from slightly different angles. The multiple images improved the accuracy of the matching. Like NIST, Australia determined through testing that Cognitec had the highest-performing solution.

The Future of Science (and Biometrics)

Dr. Eric Haseltine, Associate Director of Research for the NSA, took the audience on a whirlwind ride to the edge of modern science. A former neuro-scientist, Dr. Haseltine was with Disney before joining the NSA--top that resume if you can.

He postulated that major scientific breakthroughs have had one or many of the following bases:
- Revolutionary equipment
- Fringe theory(s)
- Convergence of disciplines
- Removal of humans from the center of the concept

For instance, the discovery of microbes was possible due to the newly-invented microscope, and the development of the modern theory of disease was based on the (then) fringe theory of germs. The understanding of the structure of DNA was the result of a convergence of disciplines, and the modern theory of evolution removes the human from the center of our concept of nature. Tomorrow's scientific breakthroughs could come from new equipment such as atomic force microscopes, or be based on a fringe theory of today such as the Gaia hypothesis.

He then extended this line of thinking to predict the future of biometrics.

Revolutionary equipment
LIDAR -- a new type of laser for long-distance collection of biometrics such as heartbest rhythm, skin characteristics, etc.
Ultrawideband radars--for non-invasive deep imaging.

Fringe theories
Magneto-neuroimaging--"reading the mind" by matching magnetic imaging of the brain to thought states such as lying, stress, etc.
IR lie detection--judging lying or stress by fine imaging of the blood flow in the face via IR imaging

Convergence of disciplines
The ultimately successful biometric system that is considered strong enough to protect our most important national secrets will likely be based on a combination of the existing and developing biometric disciplines.

Removal of humans from center
The two most popular and developed biometrics, fingerprints and face recognition, mimic the two most-used senses of humans. Future biometrics may be based on other animal skills--electromagnetic field sensing (fish), ultrasound (bats), or smell (dogs).

His grasp of the variety of subjects was familiar and pretty thorough. We asked him afterward how he tracks so many obscure fields of scientific research--"It's just part of the job."

On the Floor

Lumidigm

The company uses multiple spectrums of light to study the characteristics of the layers of skin. The fingertip sensor uses colored LEDs and imaging cells. As each LED illuminates, the sensor detects the resulting image, as well as the time it took to receive. This provides a rich set of data related to the layers of the skin.

The current application is to determine "liveness"--to tell that it is a real living person providing the fingerprints, hand geometry, or other biometric data. Customers include RSI (hand geometry biometric company), and Smith and Wesson for use in smart guns. The company is a spin-off of Insight, a medical imaging company. Kristi Nixon, chief scientist, stated that the skin-layer data is significantly unique to serve as a biometric, but for now they are focusing on liveness.

West Virginia University

WVU has just started offering a BS in biometrics. In addition the school hosts the Center for Identification Technology Reseach (CITeR)--an NSF-funded biometric research center. Current work includes multimodal biometric systems, error estimation, and socio-legal research of perceptions surrounding biometrics. With the West Virginia University programs, and the FBI center in Clarksburg, the state of West Virginia is currently at the heart of the development of the biometrics industry.

JonesCam

This is a small, pen-shaped camera that is worn on a headmount and connected via wire to a transmitter elsewhere on the body. The receiver is attached to a computing system (Wintel). The range is up to several hundred yards with omni antennae, to half a mile with high-gain directional antennae. The computing system can run face recognition or simply collect video. The visibility of the camera (it is obvious to observers) is meant to server as a visual deterrent, but the size, weight, and especially the cord made it look uncomfortable to wear. Target markets are police, military, surveillance agencies (our talk was interrupted when an NSA rep stopped by), and rescue.

BioScrypt

This company created a fingerprint recognition algorithm based on correlation--analysis of the entire image of the fingerprint--as opposed to minutae, which only considers "events" on the fingerprint ridges. The correlation system, based on waveform analysis, produces a smaller template from the data than a minutae system (384 bytes). The waveform analysis technique is patented. At the recent FVC 2002, it took 19 of 24 gold medals for both speed and accuracy. Bioscrypt provides sensors and complete recognition systems to industry an government. They currently have over 55,000 devices in operation throughout the US.

Cherry

Cherry is a keyboard and mouse producer, that offers devices with fingerprint readers built-in. The sensor is a semiconductor-based capacitive reader developed by Siemens. The devices do not come with a recognition system. They just provide a sensor at the desktop.

Technoimagia

Technoimagia is a Japanese company that produces fingerprint authentication modules for consumer or enterprise devices. The newest product is the Fp-Stick, a flash memory USB dongle that is protected by the FingerChip thermal imaging fingerprint device from ATMEL. The enrollment template and recognition engine are stored in the dongle, and are used to authenticate the owner when the device is plugged in.

Iridian

Iridian is the primary supplier of iris recognition technology to the industry. Their proprietary iris recognition system is licensed to manufacturers for bundling with iris cameras, and to system integrators, for installation. Customers (camera manufacturers) include GoldStar and Panasonic. Enrollment and challenge readings can be completed through eyeglasses or contacts, but must match--if the challenges will include eyeglasses or colored contacts, the enrollment must as well. The system has a proprietary solution for determining liveness and defeating fraud attempts.

0335.3 Semiconductors

***Demonstration of First Commercial Optical Processor by Lenslet
(October 14)

Lenslet Ltd., a developer of optical digital signal processing, today showcased live demonstrations of the company's product, EnLight, a commercial optical digital signal processor at the MILCOM exhibition in Boston, MA. The processor is specified to run at a speed of 8 Tera (8,000 Giga) operations per second, which the company states is 1000 times as fast as current DSPs. Lenslet stated that this new product will enable new applications in the fields of defense, homeland security, multimedia and communications.

Some potential benefits include enhanced communications in noisy channels, multi-channel interference cancellation, multi-protocol receiver (SDR), improved resolution and image for SAR radars, digital beam forming, enhanced signal detection in EW/RWR systems, real time, multi channel video compression and processing at high image resolutions (H.264 compression for multi HDTV channels), etc. In the area of Homeland Security: improved throughput and detection accuracy for baggage scanning and multi sensor threat analysis.

This optical processor was developed over the past three years by researchers, scientists and engineers supported by a group of international professors from the realms of optics, physics and signal processing. Lenslet's optical processor offers a combined solution of optics and silicon in the format of a standard electronic board card with standard interfaces and development tools as generally accepted in the industry.

Established in 1999 and headquartered in Herzelia Pituach, Israel, Lenslet specializes in the design, development and marketing of Optical Digital Signal Processing Engines (ODSPE).

http://www.lenslet.com


***ATI Technologies Selects NeoCircuit for Automated Circuit Sizing
(October 31)

Neolinear, Inc. today announced that ATI Technologies, Inc. has chosen NeoCircuit for analog and RF circuit sizing.

NeoCircuit, used in a design flow, sizes advanced analog, custom mixed-signal or RF circuit topologies to a set of specifications using the customer's foundry of choice and simulation environment. Neolinear stated that analog, mixed-signal and RF designers realize a 5-10x productivity improvement through NeoCircuit's adaptive analog circuit design capabilities. By capturing the design intent in the form of constraints, NeoCircuit enables IP creation of re-usable mixed-signal and RF system architectures.

ATI Technologies Inc. is a designer and manufacturer of 3-D graphics and digital media silicon solutions. Neolinear, a mixed-signal IC design automation provider, develops software and solutions for rapid design and re-use of analog, RF, mixed-signal and custom digital circuits.

http://www.neolinear.com

 

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

Copyright 2005 4th WAVE, Inc.

To subscribe to WAVE go to

http://www.wave-report.com

To unsubscribe also use the Wave Report Home page or send the preformatted UNSUBSCRIBE message:

List Management - Unsubscribe

Previous issues of WAVE, as well as other info can be found at

http://www.wave-report.com
http://www.3dlinks.com

Comments on or questions about the WAVE may be sent to:

Fourthwave Staff

or the below individuals below:

John N. Latta - Editor-In-Chief

Michael Robertson - Web Editor

The WAVE Report may be redistributed in full for individual readership and posted to newsgroups, Web, and FTP sites. This publication may not be reprinted or redistributed for profit. Short quotes are permitted but must be attributed to the WAVE Report. 4th Wave retains the copyright to the WAVE Report.