The VeriFinger Extended SDK is designed for rapid integration of fingerprint verification and identification technology into client/server or web based environments. VeriFinger Extended SDK supports development in the most popular development languages and also supports MS Windows, Linux and MAC OS X Platforms. The Extended SDK provides high quality documentation, programming samples and a pre-built matching server and client sample.

VeriFinger algorithm features and capabilities

All performance tests were made on a PC with Intel Core 2 Duo processor running at 2.66 GHz.

In 1998 Neurotechnology developed VeriFinger, a fingerprint identification algorithm designed for biometric system integrators. Since that time, Neurotechnology has released more than 10 versions of the VeriFinger algorithm, providing the most powerful fingerprint recognition algorithms to date.

The latest VeriFinger 6.1 version is NIST MINEX compliant, as it is based on the MegaMatcher fingerprint identification engine that has been certified by NIST for use in personal identity verification (PIV) program applications.

The VeriFinger algorithm follows the commonly accepted fingerprint identification scheme, which uses a set of specific fingerprint points (minutiae) along with a number of proprietary algorithmic solutions that enhance system performance and reliability. Some are listed below:

• Rolled and flat fingerprints matching. The VeriFinger algorithm matches flat-rolled, flat-flat or rolled-rolled fingerprints with high reliability, as it is tolerant to fingerprint deformations. Rolled fingerprints have much bigger deformation due to the specific scanning technique (rolling from nail to nail) than those scanned using the "flat" technique. Conventional "flat" fingerprint identification algorithms usually perform matching between flat and rolled fingerprints less reliably due to the mentioned deformations of rolled fingerprints.
• Tolerance to fingerprint translation, rotation and deformation. VeriFinger's proprietary fingerprint template matching algorithm is able to identify fingerprints even if they are rotated, translated, deformed and have only 5 - 7 similar minutiae (usually fingerprints of the same finger have 20 - 40 similar minutiae) and matches 5,000 - 14,000 flat fingerprints per second (when fingerprint image size is 300 x 300 pixels).
• Faster matching using pre-sorted database entries. For some identification tasks VeriFinger's effective matching speed can be increased to 15,000 - 70,000 fingerprints per second by pre-sorting database entries using certain global features. 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 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 correspondingly, the matching speed increases.
• Identification capability. VeriFinger functions can be used in 1-to-1 matching (verification), as well as 1-to-many mode (identification).
• Image quality determination. VeriFinger is able to ensure that only the best quality fingerprint template will be stored into database by using fingerprint image quality determination during enrollment.
• Adaptive image filtration. This algorithm eliminates noises, ridge ruptures and stuck ridges for reliable minutiae extraction – even from poor quality fingerprints – with a processing time of 0.1 - 0.2 seconds. A screenshot of the VeriFinger demo application shows an initial flat fingerprint image (left window), and the same image after the noise filtering and processing by VeriFinger (right window), with minutiae positions and directions marked by red circles and lines.
• Features generalization mode. This fingerprint enrollment mode generates the collection of generalized fingerprint features from a set of fingerprints of the same finger. Each fingerprint image is processed and features are extracted. Then the features collection set is analyzed and combined into a single generalized features collection, which is written to the database. This way, the enrolled features are more reliable and the fingerprint recognition quality considerably increases.
• Scanner-specific algorithm optimizations. VeriFinger 6.1 includes algorithm modes that help to achieve better results for the supported fingerprint scanners.

Technical Specifications and Reliability Test Results

All tests were performed on one core of Intel Core 2 Duo running at 2.66 GHz.

500 dpi is the recommended fingerprint image resolution for VeriFinger. The minimal fingerprint image resolution is 250 dpi.

All fingerprint templates should be loaded into RAM before identification, thus the maximum fingerprint templates database size is limited by the amount of available RAM.

The table below shows the technical specifications of VeriFinger 6.1 algorithm. The algorithm's performance depends on fingerprint scanner that was used for collecting fingerprint images, thus the specifications are given for two groups of flat fingerprint scanners:

• Biometric scanners in these specifications are scanners that produce fingerprint images of about 300 x 300 pixels. These scanners are usually compact and inexpensive. An example of biometric scanner is DigitalPersona U.are.U 4000.
• AFIS-class scanners in these specifications are flat fingerprint scanners that have at least 1" x 1" fingerprint sensors and produce fingerprint images of at least 500 x 500 pixels or even larger images. These scanners are mostly intended for use in large-scale AFIS projects that need to collect high quality fingerprint images. An example of AFIS-class scanner is Cross Match Verifier 300

¹ For databases with 500 or more fingerprints. Use with smaller sample fingerprint databases typically yields lower speed.

Digital Persona
U.are.U 4000

Click to zoom

Cross Match
Verifier 300 LC

Click to zoom

We present the testing results to show how VeriFinger 6.1 technical specifications correspond the practical algorithm's performance and reliability evaluations.

Flat fingerprint databases were collected with two fingerprint scanners for algorithm testing:

• DigitalPersona U.are.U 4000. 1,400 fingerprint images were collected using this scanner, with image size 318 x 330 pixels.
• Cross Match Verifier 300 LC. 1,600 fingerprint images were collected using this scanner, with image size 504 x 480 pixels.

Three tests were performed with each database:

• Test 1 maximized matching accuracy. VeriFinger 6.1 algorithm reliability in this test is shown as red curves on the ROC charts.
• Test 2 maximized matching speed. VeriFinger 6.1 algorithm reliability in this test is shown as green curves on the ROC charts.
• Test 3 minimized template size. VeriFinger 6.1 algorithm reliability in this test is shown as blue curves on the ROC charts.

Receiver operation characteristics (ROC) curves are usually used to demonstrate the recognition quality of an algorithm. ROC curves show the dependence of false rejection rate (FRR) on the false acceptance rate (FAR). Charts with ROC curves for both databases are available on the right.

VeriFinger fingerprint identification algorithm versions consistently have shown some of the best results for reliability in several biometric competitions, including the International Fingerprint Verification Competition (FVC2006, FVC2004, FVC2002 and FVC2000) and the National Institute of Standards & Technology (NIST) Fingerprint Vendor Technology Evaluation (FpVTE 2003), where Neurotechnology ranked among the top five companies for accuracy in single-finger tests.

Technology Awards

• MINEX certification
• FVC2006 results
• FpVTE 2003 results
• FVC2004, FVC2002 and FVC2000 results
• Comments on competitions' results
• NIST Proprietary Fingerprint Template (PFT) testing

MINEX Certification

In 2007 MegaMatcher SDK fingerprint technology received full MINEX Certification. NIST certified MegaMatcher for use in personal identity verification program applications. MegaMatcher fingerprint technology is also used in VeriFinger SDK.

The Minutiae Interoperability Exchange Test (MINEX) evaluates fingerprint template encoding and matching to determine compliance with the government's Personal Identity Verification (PIV) program for the identification and authentication of Federal employees and contractors. The MINEX program provides measurements of fingerprint algorithm performance and interoperability to both government and commercial entities.

MegaMatcher is one of only 12 algorithms worldwide to receive full MINEX certification for both fingerprint template encoding and matching. This certification puts MegaMatcher SDK into the U.S. government buyers' certified list of fingerprint recognition algorithms.

Fingerprint Verification Competition (FVC2006)

Neurotechnology is pleased to announce that our results in the Fingerprint Verification Competition (FVC2006) achieved the highest ranking when using the most realistic benchmark for real-world biometric applications, "Average Zero FMR."

Neurotechnology participated in FVC2006 under the name Neurotechnologija. In 2008 the company changed it's corporate name to Neurotechnology.

FVC2006 Open Category results.
The whole page is available at the FVC2006 web site.
Neurotechnology algorithm is denoted there as P058.

Neurotechnology also won four gold medals, two silver and two bronze medals in the FVC2006 Open Category.

Our algorithm took second place in the FVC2006 Light Category, according Average Zero FMR benchmark. The algorithm won one gold and four bronze medals in this category.

Considering Competition Results in Real-World Applications

For each participating algorithm, the Fingerprint Verification Competition (FVC2006) measured several reliability parameters, including:

• EER (Equal Error Rate) – where the False Acceptance Rate (FAR) is equal to the False Rejection Rate (FRR),
• FMR 100 (FRR at the FAR=1% level),
• FMR 1000 (FRR at the FAR=0.1% level),
• Zero FMR (FRR at the FAR=0% level).

When considering the results of competitions, it is important to put the competition criteria into the perspective of real-world biometric applications.

The goal of many real-world applications of biometric technology is to let the "good guys" in while keeping the "bad guys" out. In most security situations, keeping a few of the "good guys" out is more acceptable than letting a few "bad guys" in. Thus, most real-world applications of biometric technology are set to have a low FAR. Most real applications set the FAR as close to zero as possible. A FAR=0.001% is common and sometimes FAR=0.0001% or even less are used. This minimizes the number of people who are incorrectly accepted into the system (or allowed entry). When the FAR is low, the FRR is higher, which means the system may incorrectly refuse entry to someone who should be there. A more reliable algorithm means you will have a lower FRR when the FAR is very low (near to zero).

In this sense, other than EER, which represent reliability in very high FAR area only, the Zero FMR rate is the most adequate benchmark for evaluating real-world biometric applications.

The Fingerprint Vendor Technology Evaluation (FpVTE 2003)

Conducted by the National Institute of Standards & Technology (NIST) on behalf of the Justice Management Division (JMD) of the US Department of Justice

Neurotechnology participated in FpVTE 2003 under the name Neurotechnologija. In 2008 the company changed it's corporate name to Neurotechnology.

Neurotechnology's algorithm achieved one of the best reliability results in the Middle Scale Test among FpVTE 2003 participants:

• In real-world scenarios, Neurotechnology's algorithm would show even higher accuracy levels.
• See FpVTE web site for a detailed report of the evaluation results^*.

^* Results shown from the NIST FpVTE 2003 do not constitute endorsement of any particular system by the government.

FVC2004, FVC2002 and FVC2000 results

Organized by Biometric Systems Lab (University of Bologna), Pattern Recognition and Image Processing Laboratory (Michigan State University) and the Biometric Test Center (San Jose State University)

Neurotechnology participated in FVC2004, FVC2002 and FVC2000 under the name Neurotechnologija. In 2008 the company changed it's corporate name to Neurotechnology.

Neurotechnology's algorithms consistently showed some of the best reliability results among participants, earning the following awards:

• FVC2004 (See FVC2004 web site for details)
  • Open Category: four gold, three silver and two bronze medals for the VeriFinger algorithm
  • Light Category: one gold, six silver and three bronze medals for the FingerCell algorithm
• FVC2002 (See FVC2002 web site for details)
  • One silver and two bronze medals
• FVC2000 (See FVC2000 web site for details)
  • VeriFinger algorithm showed the best reliability results among all participants.

Since the FpVTE 2003 and FVC2004 competitions were held, Neurotechnology has developed many algorithm improvements on the versions tested in the contests (both algorithms were submitted in 2003). New fingerprint filtration functions were developed, allowing better filtration of low quality images. Additionally, the generated templates size has been decreased from 300 - 600 bytes to 150 - 300 bytes per fingerprint by using features set optimization. Also, identification speed has been increased from 5% to 100%, depending on the number of fingerprint minutiae. All these improvements allow us to achieve even better results in our products.

Comments on competitions' results

The FpVTE protocol was strict and did not allow using some of our advanced algorithm features, which, in a real world application, would further increase the recognition quality. Particularly, the MST set contained images from different scanners, but each certain image scanner model was not disclosed. In a real world scenario, specific parameters would be set for each specific scanner type. This would allow the algorithm to perform at an even higher accuracy level.

Another such real world example that was not simulated in the FpVTE protocol is the ability to generate globalized or generalized features templates by capturing several images from the same finger and combining the templates into a single features set. Using a generalized features set can significantly improve the algorithm's reliability and produces improved matching scores. In the FpVTE MST set such a method could not be used, as only two matched fingerprints were allowed for consideration.

The FVC protocol is very useful for comparing different vendors' algorithms, however it only allows comparison of verification (1-to-1 matching) but not identification (1-to-many matching) results. One of the strongest capabilities of Neurotechnology's algorithms is fast reliable identification, therefore a 1-to-many test would better reflect our real algorithm ranking among the participants.

FVC uses databases that are not from real applications (more information), but rather uses fingerprint sets which had been specially collected for the competition (some with certain distortion or noises highlighted). In this way, various distortion and noise statistics of the fingerprints did not correspond to real world application statistics, and vendors' results may be not completely adequate to apply to real life situations.

Like the FpVTE, the FVC did not allow us to generate globalized or generalized features templates by capturing several images from the same finger and combining the templates into a single features set. Using a generalized feature set can significantly improve the algorithm's reliability and produces improved matching scores. In the FVC such a method could not be used, as information from only two matched fingerprints was allowed for consideration.

NIST Proprietary Fingerprint Template (PFT) Testing

Since June 2003, NIST has been conducting tests of fingerprint-based biometric matching systems using vendor supplied SDKs. The main result obtained from these evaluations is an estimate of how well commercial products performed one-to-one matching for verification over a wide range of fingerprint image qualities.

Unlike the Fingerprint Vendor Technology Evaluation (FpVTE), these evaluations are ongoing and new SDKs can be included in the test at any time.

The latest MegaMatcher SDK was tested in NIST PFT Testing and the results are available under "1T" label.