.Products\ TigerAFIS

Click to download TigerAFIS brochureDownload
TigerAFIS Brochure (1.8MB)

Automated fingerprint identification is the process of automatically matching one or many unknown fingerprints against a database of known and unknown prints. Tiger IT is one of the few companies worldwide to develop an NIST certified AFIS, and the first among the South Asian countries. Tiger IT is ranked number 1 in the Minutiae Interoperability Exchange (MINEX) test by the NIST (National Institute of Standards & Technology). According to the NIST MINEX benchmarking, TigerAFIS matcher showed False Non Match Rate of 0.0002 with self-generated templates along with a fixed False Match Rate of .01. [Source: http://www.nist.gov/itl/iad/ig/ominex_test-results.cfm]. This makes Bangladesh the producer of the world’s best AFIS and the only South Asian country to develop a NIST certified AFIS. Tiger IT is also a provider of pre-packaged and customized e-business solutions and strategic consultancy.

TigerAFIS uses specific components of a fingerprint template at various stages of the matching process. The industry leading matching process takes place in two major levels: (a) Index Matching, and (b) Exact Matching. The Index Matching is an intelligent way to quickly filter a range that will be used later to perform an Exact Matching operation of the shortlisted fingerprints. Fingerprint image is a complex data object that is subject to multiple noise sources. For years, we’ve looked into the implementation and use of our products in different geographic locations worldwide, acquiring the means to understand not only why quality issues are seen on collected biometric data, but, more importantly, how they arise. TigerAFIS defines quality as a complex measurement derived from intensity, intensity variation, values for angular consistency, and other features. It is computed at every point of the image. The system is flexible enough and allows different classifications for quality, depending on the task at hand.

Tiger IT AFIS provides fingerprint technology and products that meet the following relevant standards:

  • Finger Image-Based Data Interchange Format
    • INCITS 381-2004, American National Standard for Information Technology
    • ISO/IEC 19794-4, International Organization for Standardization
  • Fingerprint Image Quality
    • NISTIR 7151 - NIST Interagency Report, August 2004
  • Finger Minutiae Format for Data Interchange
    • INCITS 378-2004, American National Standard for Information Technology
    • ISO/IEC 19794-2, International Organization for Standardization
  • Data Format for the Interchange of Fingerprint, Facial, & Scar Mark & Tattoo (SMT) Information
    • ANSI/NIST ITL 1-2007 – NIST Special Publication 500-271
    • ANSI/NIST ITL 2-2008 - NIST Special Publication 500-275
  • Electronic Fingerprint Transmission Specification
    • IAFIS-DOC-01078-8.1 (V8.1) – Criminal Justice Information Services, Federal Bureau of Investigation, Department of Justice

Templates

A template contains information on the features that are required to uniquely identify a fingerprint. Extraction Fidelity is an important concept that TigerAFIS engineers value. From our experience, analysis of multiple publications, and from many independent sources and expert reviews, we conclude that our image enhancement is one of the best in the world. We provide considerable importance to the accuracy with which a template, feature set, or model is generated from the sample. This extraction fidelity is directly related to the accuracy of TigerAFIS’s feature extraction algorithm.

SCREENSHOTS
Click to enlarge Click to enlarge

TigerAFIS creates templates that have reasonably smaller file size with sufficient information on features, compared to many competitors. On average, the templates for each fingerprint are about 1.3KB in size. Our AFIS can generate approximately two templates per second. It also extracts specific information on fingerprint images that are stored as Index, which is about 410 bytes on average for each fingerprint. Indexes allow the matcher to perform initial short-listing by using fewer details.

Server-SideTemplate Generation: TigerAFIS has an automated template generation feature. This application component may operate at the central data center, and is especially important if a program has existing biometric images in the database that needs to be processed to generate ANSI/ISO standard templates for storage on a credential (for subsequent verification).

Fingerprint Quality Scores for Templates: TigerAFIS assigns an image quality score for each fingerprint recorded. It logs the scores of all the fingerprint images during the creation of the templates from the WSQ images. These scores can be used to identify poor quality fingerprint and track the corresponding person ID.

Adjudication

Adjudication is the process to determine if a hit in the AFIS is in fact a real or a false match. The TigerAFIS system includes an adjudication application where all hits are stored in a message queue. The adjudication stations can run the TigerAFIS adjudication software, where operators can manually determine if a hit is a real or a false. The adjudication workflow options include the ability to randomly assign a hit result to an adjudicator (given multiple adjudicators exist in the enterprise), and to have more than one adjudicator review each hit result (again this can be randomly assigned to the second adjudicator to prevent any collision).

SCREENSHOT
Adjudication

In addition, an “Investigate” state can be instantiated to enable a committee or assigned investigator to review the file in much more detail and scrutiny, including possibly calling the person back in for an interview. In general, the steps to investigate / verify an irregular transaction are illustrated below. This can be customized based on the application workflow and operational requirements as follows:

  • Upon detection of duplicate registration by the ID server, the subject (applicant) and the duplicate registration are available to the adjudication application.
  • The adjudication application retrieves the demographic data, photograph and fingerprint data from the biometric database.
  • Display the subject and duplicate registration side-by-side on the screen for comparison and verification.
  • Perform visual investigation (e.g. view fingerprint, see matched and unmatched minutiae, etc.).
  • Approve or reject the transaction.
  • Print investigation report if desired.

Administration

TigerAFIS is delivered with web services that support a variety of automated functions including, but not limited to:

  • Enroll - Check for existing aliases (duplicates) in the AFIS, and if none is found insert new record into the AFIS. If duplicate is found it returns the previously enrolled matching record for adjudication.
  • Update – Replace previously enrolled record in the AFIS with the new record for same person after verifying fingerprints to ensure it is the same person.
  • Identify – Search the record against all previously enrolled records and return the results (match or no-match).
  • Verify – Compare the record against a designated previously enrolled record (using the unique record locator number) and return the results (match or no-match).
  • Delete – Remove previously enrolled record in the AFIS.
SCREENSHOT
Regional Scrubing

Additional lower level functions for more specific control and configuration settings are also available, which can be set in each transaction request itself, or on a global basis including, but not limited to:

  • Set Match Threshold(s) – Adjust one or more match thresholds away from the default settings to adjust the accuracy of the system.
  • Set Fusion Settings – Adjust how matching results from multiple fingers used in a transaction are fused to generate a final match score.
  • Set Identify Settings – Adjust various parameters that determine how fast and accurate search transactions occur.
  • Set Enroll Settings – Adjust various settings that determine how fast and accurate enrollment transactions occur and whether to mark and retain alias records in the database (primary vs. alias records).
  • Administration and Reporting – Adjusts various settings regarding logs, audit trails, reports and administration settings.
  • Database Settings – Adjusts where and how the images and templates are permanently stored.
  • Failover Settings – Adjusts how failure conditions are managed including options to suspend processing while the failure is recovered, continue operations ignoring the failed matcher, redistribute templates loaded on the failed matcher to all other matchers of the same finger and continue processing.
  • Additional Administrative features:
    • User Management
    • Person Query
    • Matcher Status Monitoring
    • Node Status Monitoring
    • Scrubbing / De-duplication and Server Load Status Monitoring
    • Scrubbing / De-duplication Information Monitoring
    • Database Status Information Monitoring

Benchmark

In order to evaluate fingerprint recognition algorithms using vendor-supplied AFIS Software Development Kits, NIST performs a rigorous testing method to calculate accuracy of every fingerprint recognition algorithm in different databases. In the most recently completed NIST PFT Test (June 2003 – February 2010) a total of 55 algorithms from world’s top AFIS vendors were used.

Details available at: http://www.nist.gov/itl/iad/ig/pft_2003.cfm
  • #1 for Lowest EER in FVC-OnGoing (as of Aug ’11)
  • #1 in NIST MINEX 2009
  • Top 3 in NIST PFT Tests in US POE / POEBVA Datasets
  • Highly scalable architecture
  • One of the best template generators
  • User friendly adjudication interface
  • High flexibility for system administration, tweaking, and enhancement
  • Proven record on working with some of the largest biometrics projects
Rankings
Tiger IT achieves lowest FRR at FAR
of 1% in MINEX Test by NIST in 2009 - Click to enlarge TAR at FAR of 1% (DOS Database) - Click to enlarge TAR at FAR of .01% (POE Database) - Click to enlarge Overall EER (POE Database) - Click to enlarge Overall EER (POEBVA Database) - Click to enlarge

Hardware

Tiger IT Automated Fingerprint Identification System (TigerAFIS) is a highly scalable fingerprint matching array that enables authorities to prevent multiple registrations by any one individual leveraging one or more aliases. TigerAFIS is a set of services that can reside on a single server, or distributed across an array of servers, depending on the anticipated transaction volume. Leveraging n-tier architecture, Tiger IT AFIS can scale to hundreds of multi-core servers all operating in parallel to perform high-speed fingerprint identification. The fundamental tier is the ‘matcher’ service. In large programs, the Tiger IT AFIS matchers are commercial off the shelf (COTS) servers running a Linux operating system that require no proprietary hardware to house and search all of the fingerprint templates in high speed RAM. Each matcher houses a subset of the overall fingerprint database; the persons, with all their fingers, get evenly distributed between the servers. For example, if there are 10M persons and 10 servers, each server will have 1M persons’ records. The next tier up is the ‘matcher controller’ service, also operating on dedicated Linux and COTS servers in large programs, which moderates all enrollment and search request transactions coming from the Tiger IT AFIS Web Service (called by Tiger IT AWSE and other Tiger IT or any other systems and applications). The matcher controller organizes enrollment and search requests and passes them to the matchers for processing. The matcher controllers also collate and fuse all results returned by the matchers and present either a final match or a candidate list back to the Tiger IT AWSE and/or application making the request. Each matcher controller can moderate multiple matchers. The number of matchers moderated by each matcher controller depends on the anticipated transaction volumes of the program.

TigerAFIS is built on Intel Xeon processor based servers. Each matching server (each server typically supports multiple cores) is capable of a multi-threaded searching speed of more than 30 million matches per second without any enrolled images being ignored during the search process. Specific local and global features of the image are used to compare the fingerprints at high speed to generate a candidate list, which is reduced to any final matches using a detailed comparison.