FaceCell, Embedded Face Recognition Technology

• Reliability. The FaceCell technology is intended for hardware with lower computational capabilities than PCs.
• Identification ability. FaceCell is designed for verification (1:1 matching) and identification (1:N matching). The algorithm is able to match up to 3,000 faces per second.
• Easy integration - Easily integrated into handheld or embedded devices with built-in video cameras.
• Portability. FaceCell Embedded Development Kit is designed for easy implementation into very various and specific applications. The algorithm's ANSI C source code can be ported to various platforms and hardware.
• Embedded and PC-based multi-biometric capable technologies from the same vendor. Combined with our other technologies, FaceCell could be used in developing these advanced systems:
  • Multi-biometric embedded systems, using FaceCell EDK together with FingerCell EDK.
  • Mixed embedded/PC systems, using FaceCell EDK together with VeriLook Standard SDK or VeriLook Extended SDK.
  • Complex multi-biometric embedded/PC systems, using a combination of FaceCell EDK, FingerCell EDK, VeriLook SDK, VeriFinger SDK and VeriEye SDK.

Algorithm

The FaceCell algorithm is similar to the VeriLook algorithm and includes these features:

• Fast and accurate face localization for reliable detection of multiple faces in the images.
• Simultaneous multiple face processing and identification in a single frame. All faces in the current frame are detected in about 1 second* and then each face template is extracted in about 1 second*.
• Face quality threshold can be used during face enrollment to ensure that only the best quality face template will be stored into database.
• The FaceCell face template matching algorithm compares 3,000 faces per second*.
• Applications implemented using FaceCell EDK can handle large face databases, as one facial feature template is only 2.3 Kbytes.
• Features generalization mode generates the collection of the generalized face features from several images of the same subject.

* All performance evaluations were performed using a HP iPAQ Pocket PC with XScale PXA270 processor running at 416 MHz

Related products

• FaceCell 1.1 Library EDK
• FaceCell 1.1 source code EDK

VeriLook Face Recognition Technology

• Reliability. The VeriLook 3.2 algorithm has been tested with standard face databases (FERET, XM2VTSDB and others). The results are one of the best among existing face identification systems on the market.
• Speed. VeriLook 3.2 face enrollment time is less than 0.3 sec. and matching speed is 100,000 faces per second in 1:N identification mode.
• Live face detection. VeriLook is able to prevent cheated by placing a photo of another person in front of a camera. this kind of security breach by determining whether a face in a video stream belongs to a real human or is a photo.
• Multiple Face Detection
  
  
  
  Multiple face processing. VeriLook 3.2 detects all faces in the current frame and can process all of them simultaneously.
• VeriLook doesn't require any specific hardware. The face image can be obtained from a webcam or other low cost camera. Image processing and recognition are performed on an ordinary PC.
• Biometrical template record can contain multiple face samples belonging to the same person.
• Fingerprint, iris and face recognition technologies from the same vendor.
• VeriLook is designed not only for verification, but also for identification (1:N matching).

Related Product

• VeriLook 3.2 Standard SDK is intended for PC-based biometrical application development. It includes Matcher and Extractor components, programming samples and tutorials, camera manager library and software documentation. The SDK allows the development of biometric applications for Microsoft Windows, Linux or Mac OS X operating systems.
• VeriLook 3.2 Extended SDK is intended for biometrical Web-based and network application development. It includes all features of Standard SDK. Additionally, the SDK contains sample client applications, tutorials and a ready-to-use matching server.

VeriLook Matching Server runs on PCs with Microsoft Windows or Linux and includes these additional components:

• Server administration tool for monitoring servers state, managing log and other server administration tasks;
• Support modules for MySQL and Oracle databases.
• Sample client applications:
  • C# sample (for Microsoft Windows);
  • C sample (for Linux);
  • Sample Java applet.

VeriEye Iris Recognition Technology

• Reliability. VeriEye 2.0 algorithm shows excellent performance when tested on all publicly available datasets
• Speed. Enrollment time is less than 0.5 sec. and matching speed is configurable 50,000-150,000 irises per second in 1:N identification mode.
• Uniqueness. The new proprietary iris recognition algorithm is based on original methods that solve the drawbacks and limitations of existing state-of-the-art algorithms.
• Robustness. Eye irises are detected even when the images have obstructions, visual noise and different levels of illumination.
• Simple multi-biometric system integration. Compatibility with fingerprint and facial identification technologies from the same vendor.
• Flexible licensing and pricing. VeriEye is offered for a competitive price. Developers can select from several types of SDK and licensing models.

Algorithm

The VeriEye 2.0 iris recognition algorithm implements advanced iris segmentation, enrollment and matching using robust digital image processing algorithms:

• VeriEye uses active shape models that more precisely model the contours of the eye, resulting in correct iris segmentation when perfect circles fail.
• Correct segmentation is achieved even when the centers of the iris inner and outer boundaries are different (see Figure 1). The iris inner boundary and its center are marked in red, the iris outer boundary and its center are marked in green.
• Correct segmentation when iris boundaries are definitely not circles and even not ellipses (see Figure 2) and especially in gazing-away iris images.
• Even when iris boundaries seem to be perfect circles, recognition quality can still be improved if boundaries are found more precisely (see Figure 3). Note these slight imperfections when compared to perfect circular white contours.
• Automatic interlacing detection and correction results in maximum quality of iris features templates from moving iris images.
• Elimination of lighting reflections, eyelids and eyelashes obstructions.
• Detection and correction of gazing-away iris images (see Figure 4). A gazing-away eye is correctly segmented and transformed as if it were looking directly into the camera.
• Configurable matching speed varies from 50,000 to 150,000 comparisons per second. The highest speed still preserves nearly the same recognition quality

These parameters were determined for one core of Intel Core 2 Duo running at 2.66 GHz

VeryEye SDK

• VeriEye 2.0 Standard SDK is intended for PC-based biometrical application development. It includes Matcher and Extractor components, programming samples and tutorials, eye iris scanner drivers and software documentation. The SDK allows the development of biometric applications for Microsoft Windows, Linux or Mac OS X operating systems.
• VeriEye 2.0 Extended SDK is intended for biometrical web-based and network application development. It includes all features of Standard SDK. Additionally, the SDK contains a ready-to-use matching server

VeriEye Matching Server runs on PCs with Microsoft Windows or Linux and includes these additional components:

• Server administration tool for monitoring servers state, managing log and other server administration tasks;
• Support modules for MySQL, Oracle, MS SQL Server and SQLite databases.
• Sample client applications:
  • C# sample (for Microsoft Windows)
  • C sample (for Linux)

Introduction

Nowadays the need for automated biometrical identification systems is increasing in civil and forensic fields of applications. The fast and accurate identification becomes particularly critical for large-scale applications, such as passport and visa documentation, border crossings, election control systems, credit card transactions control and crime scene investigations. Many countries, including the US, European countries and others incorporate biometrical data into passports, ID cards, visas and other documents for using in large national scale automatic biometrical identification systems.

Automated fingerprint identification systems (AFIS) have been widely used in forensics for the past two decades, and recently they became relevant for civil applications. Whereas large-scale biometrical applications require high identification speed and reliability, multi-biometric systems that incorporate both face and fingerprint recognition offer a number of advantages for improving identification quality and usability.

Large-scale automatic biometrical identification systems have a number of special requirements, which are different from those for small or middle scale biometrical systems:

• The system must perform reliable identification with large databases, as biometrical identification systems tend to accumulate False Acceptance Rate with database size increase and using single fingerprint or face image for identification task becomes unreliable for large-scale application. Several biometrical samples should be used to increase identification reliability, and multi-biometrical technologies (i.e. collecting fingerprint and face samples from the same person) are often employed there for additional convenience.
• The system must show high productivity and efficiency, which correspond its scale:
  • System scalability is important, as the system might be extended in the future, so high productivity level should be kept by adding new units to the existing system.
  • Daily number of identification requests could be very high.
  • Identification request should be processed in a very short time (ideally – in real time), thus high computational power is required.
  • Support for large databases (tens or hundreds millions of records) is required.
  • General system robustness. The system must be tolerant to hardware failures, as even temporary pauses in its work may cause big problems taking into account the application size.
• The system must support major biometrical standards. This should allow using the system generated templates or databases with the systems from other vendors and vice versa.
• The system must be able to match flat (plain) fingerprints with rolled fingerprints, as many institutions collect rolled fingerprint databases.
• The system must be able to work in the network, as in most cases client workstations are remote from the server with the central database.
• A forensic system must be able to edit latent fingerprint templates in order to submit latent fingerprints into AFIS for the identification.

Despite all these requirements, the system price should be as low as possible. Many existing AFIS are specialized for criminalistics or other particular applications and are quite expensive. Neurotechnology offers a technology for large-scale AFIS and multi-biometric face-fingerprint identification products, which meets all of the requirements mentioned above.

Why MegaMatcher?

Neurotechnology has experience in collaborating with many biometrical system integrators, who develop large-scale biometrical systems. To address their requirements, the company has developed the MegaMatcher multi-biometrical technology, intended for large-scale face-fingerprint systems and AFIS integrators. MegaMatcher has a set of specific features, which make it very attractive for large-scale biometric system developers:

• Multibiometrics. MegaMatcher includes fingerprint and facial recognition engines and allows integrators to use fused algorithm for better identification results or any of these engines separately. Identification reliability is a very important requirement for a large-scale system, thus usually two or more different biometrical samples from the same person are used to increase recognition reliability. Using MegaMatcher 2.1's multi-biometric technology, developers and integrators can create systems where both face and fingerprint can be scanned at the same time using inexpensive hardware, such as a fingerprint scanner and a simple webcam or photo scanner.
• Reliability. As MegaMatcher uses fusion of facial and fingerprint recognition results, the identification reliability is very high even when using large databases with millions of records. Receiver operating characteristic (ROC) curves show the reliability results for MegaMatcher 2.1. The chart compares MegaMatcher 2.1 face identification engine reliability , fingerprint identification engine and the fused face-fingerprint algorithm. These ROCs show that large-scale automated biometrical identification system based on MegaMatcher provides high identification reliability when using fingerprints, and using multi-biometrical identification results in significant reliability increase, allowing to reach almost 0% FRR.
• Matching speed. MegaMatcher is able to match up to 400,000 templates per second running the fused algorithm on a stand-alone PC with 3GHz CPU. MegaMatcher's facial recognition engine is able to match up to 500,000 faces per second, and the fingerprint recognition engine matches up to 60,000 fingerprints per second. The matching speed could be significantly increased by using the PC-based cluster.
• MegaMatcher includes cluster software for performing parallel matching, which allows to reach high performance, high availability and efficiency:
  • The effective matching speed increases proportionally to the number of the cluster's nodes and can be scalable to achieve the necessary system performance. For example, a cluster-based multi-biometrical identification system with 10 nodes is able to match up to 4,000,000 records per second, a cluster with 100 nodes - up to 40 millions records per second etc. Such scalable architecture allows to keep up the fast system response if its size becomes larger.
  • A large number of identification requests could be processed by the cluster-based multi-biometrical system. Suppose, there is a database with 10 million records. A cluster with 10 nodes (PCs with 3GHz CPU) will be able to process about 30,000 requests per day with the given database, a cluster with 20 nodes – about 60,000 requests per day and so on.
  • Fast request processing. The scalable cluster architecture for automated biometrical identification system allows to achieve real-time processing of the identification request.
  • The cluster is able to handle databases of a practically unlimited size.
  • Computer cluster is fault-tolerant, so in case of a cluster node fault, the matching speed slightly decreases, but the cluster's work remains uninterrupted.
• MegaMatcher supports BioAPI 2.0 (ISO/IEC 1978-1:2006) and other biometrical standards:
  • ISO/IEC 19794-2 (Information technology – Biometric data interchange formats – Part 2: Fingerprint minutiae data)
  • ISO/IEC 19794-4 (Information technology – Biometric data interchange formats – Part 4: Finger image data)
  • ISO/IEC 19794-5 (Information technology – Biometric data interchange formats – Part 5: Face image data)
  • ANSI INCITS 378-2004 (Finger Minutiae Format for Data Interchange)(ANSI378)
  • ANSI INCITS 381-2004 (American National Standard for Information Technology – Finger Image-Based Data Interchange Format)
  • ANSI INCITS 385-2004 (American National Standard for Information Technology – Face Recognition Format for Data Interchange)
  • ANSI/NIST-ITL 1-2000 (Data format interchange of Fingerprint, Facial, and Scar Mark and Tattoo (SMT) Information) (AN2K)
  
  Therefore, MegaMatcher fingerprint templates could be exported to another identification system and vice versa. Additionally, MegaMatcher supports WSQ fingerprint image storage format.
• The technology allows to match rolled and flat fingerprints between themselves. Usually conventional "flat" fingerprint identification algorithms perform matching between flat and rolled fingerprints less reliably due to the specific deformations of rolled fingerprints. MegaMatcher allows matching of flat-flat, flat-rolled or rolled-rolled fingerprints with high reliability.
• MegaMatcher includes network support, as components of MegaMatcher are intended to be distributed on the network.
• Effective price/performance ratio. MegaMatcher uses a PC and can work with Microsoft Windows and Linux operating systems. This configuration provides the most price/performance effective computational units for all components of the system. Therefore, developing with MegaMatcher SDK means that the system price will be reasonable for both software and hardware parts.
• MegaMatcher is fully compatible with other Neurotechnology's products: VeriFinger VeriLook,FingerCell and FaceCell.

Algorithm

MegaMatcher includes facial and fingerprint recognition engines and allows to use the new fused algorithm for fast and reliable identification in large-scale systems. Face or fingerprint identification algorithms can be used alone to develop an automated facial identification system or an AFIS respectively. Both biometrical software engines contain many proprietary algorithmic solutions, which are especially useful for large-scale identification problems. These solutions were specially developed for MegaMatcher, and some were inherited from the VeriFinger and VeriLook algorithms. Some of these solutions are listed below for each biometrical identification engine.

MegaMatcher fingerprint identification engine

• Full MINEX Certification. NIST has certified MegaMatcher fingerprint technology for use in personal identity verification program applications.
• MegaMatcher includes fingerprint image quality determination which can be used during enrollment to ensure that only the best quality fingerprint template will be stored into database.
• Template generalization is used to generate a better quality template from several fingerprints. Better quality templates result in higher identification quality.
• MegaMatcher is tolerant to fingerprint translation, rotation and deformation. It uses a proprietary fingerprint matching algorithm, which identifies fingerprints even if they are rotated, translated and have deformations.
• MegaMatcher algorithm is able to match rolled fingerprints, flat fingerprints, and also rolled with flat between themselves. Due to the specific scanning technique (rolling from nail to nail) rolled fingerprints usually have much bigger deformation than those scanned using the "flat" technique. MegaMatcher matches rolled fingerprints very well, as it is tolerant to fingerprint deformations.
• MegaMatcher can use database entries which were pre-sorted using certain global features and matches about 60,000 fingerprints per second using the pre-sorted records. Fingerprint matching is performed first with the database entries having global features most similar to those of the test fingerprint. If matching within this group yields no positive result, then the next record with the most similar global features is selected, and so on until the matching is successful or the end of the database is reached. In most cases there is a fairly good chance that the correct match will be found at the beginning of the search. As a result, the number of comparisons required to achieve fingerprint identification decreases drastically, and the effective matching speed increases correspondingly.
• Adaptive image filtration algorithm allows to eliminate noises, ridge ruptures and stuck ridges, and extract minutiae reliably even from poor quality fingerprints, with processing time of less than 1 second (all times are given for one core of Intel Core 2 Duo running at 2.6GHz).

MegaMatcher facial identification engine

• Template generalization is used to generate a better quality template from several face images. Better quality templates result in higher identification quality.
• MegaMatcher has certain tolerance to face posture that assures face enrollment convenience: rotation of a head can be up to 10 degrees from frontal in each direction (nodded up/down, rotated left/right, tilted left/right).
• Reliable face detection assures convenient face enrollment from cameras, webcams and especially various scanned documents: faces will be found on scanned pages from passports, files etc. Multiple faces can be also detected on an image and simultaneously processed.
• Live face detection. A conventional face identification system can be easily cheated by placing a photo of another person in front of
• a camera. MegaMatcher is able to prevent this kind of security breach by determining whether a face in a video stream belongs to a real human or is a photo.
• Biometrical template record can contain several face samples belonging to the same person. These samples can be enrolled from different sources and in different time thus allowing to improve matching quality. For example a person could be enrolled with and without eyeglasses or with different eyeglasses, with and without beard or moustache, etc.