Data processing: measuring – calibrating – or testing – Measurement system – Performance or efficiency evaluation
Reexamination Certificate
2000-08-17
2003-05-20
Hoff, Marc S. (Department: 2857)
Data processing: measuring, calibrating, or testing
Measurement system
Performance or efficiency evaluation
C702S019000, C702S032000, C702S108000, C702S124000, C702S179000
Reexamination Certificate
active
06567765
ABSTRACT:
BACKGROUND
1. Technical Field
This disclosure relates to finger print recognition, and more particularly, to a system and method for evaluating for fingerprint systems, which is capable of predicting performance with a high level of confidence without the need to acquire a large number of testing images.
2. Description of the Related Art
Fingerprint may be performed electronically to verify a person's identity for different applications, for example, entry into a secured area, access to a bank account, etc. Referring to
FIG. 1
, a fingerprint verification system
10
typically includes the following components. A sensor
12
is included that acquires fingerprint images, often at a resolution of about 500 pixels per inch (ppi). A feature extraction module
16
converts image pixels into a small set of characteristic features for concise image representation. The most commonly used features are the minutiae of the fingerprint; i.e. the ridge ending or bifurcation points. Other features include cores and deltas, ridge count between minutia pairs, and ridge width. Module
16
may also include a quality control sub-module
14
, which can provide feedback to users on poorly acquired images or non-fingerprint images. A fingerprint matcher
18
assigns a similarity score between a search (candidate) print and a reference print from a database
20
, and decides whether to declare a match between the pair. Normally the matcher relies entirely on the features provided by the feature extractor
16
.
In block
30
, components are provided for training system
10
to associate a given person to their fingerprint. Fingerprint data is stored in database
20
and employed by matcher
18
to later verify users stored in database
20
. Matcher
18
is part of an authentication module
32
.
In
FIG. 1
, the verification algorithm of the system
10
is divided into two modules: feature extractor
16
and matcher
18
. This separation is important in the context of performance evaluation. It permits analysis of matcher performance in isolation from the more sensor-dependent feature extractor. The matcher is nearly sensor independent. In spite of their intimate inter-relationship, these two modules tend to be impacted differently by factors affecting the performance of a fingerprint system. For instance, image warping due to finger elasticity affects exclusively the matcher
18
, while intensity shift caused by moisture in fingers affects mostly the feature extractor
16
. Of course, errors made by the feature extractor
16
are propagated through the matcher
18
and their effects need to be analyzed as well.
Since fingerprint systems are subject to strong statistic errors and the complex biometric features used for matching generally cannot be accurately described by mathematical models, it is very important to expose the systems to rigorous tests to assess their performance during development. A common method for system evaluation and validation is to use large-scale field tests. While this approach is effective, it is very costly and time consuming. Alternatively, existing fingerprint databases may be used to evaluate the algorithm portion of the system, e.g., the feature extractor
16
and the matcher
18
; however, the testing results are often skewed due to at least the reasons listed below:
1. The characteristics of the sensor from which the database was constructed are often different from the sensor in the system under evaluation. As a result, the performance of the feature extractor, usually tuned to a particular sensor for optimal performance, cannot be subjectively and realistically assessed.
2. Even if the feature extractor were tuned for images in the database, the performance of the matcher is still biased by characteristics of the statistical variability and distortion of images in the database. Performance degradation due to poorly extracted features often cannot be easily separated from inherent deficiencies of the matcher.
Another option for system performance evaluation is to use synthetic fingerprint images, for example, synthetic fingerprint images generated by OPTEL, LTD. software, e.g., Fingerprint Synthesis™. However, there exist severe limitations in the usefulness of synthetic fingerprint images. The most severe and also hardest to overcome is the extreme difficulty of generating synthetic images that can realistically mimic the characteristics of defects, artifacts and noise naturally present in real fingerprint images. As a result, the performance of a feature extractor in synthetic images tends to be a poor predictor for its performance in real images. In addition, the natural distribution of finger features is very complex and cannot be easily characterized by simple statistical models. For instance, random distribution of minutiae tends to produce optimistic estimates of FAR (false acceptance rate) and FRR (false rejection rate) distributions.
Therefore, a need exists for an evaluation technology for fingerprint systems, which is capable of predicting system performance with high confidence without the need to acquire a large number of testing images. A further need exists for a method for evaluating fingerprint systems which is accurate, quick and economical.
SUMMARY OF THE INVENTION
A system and method for evaluating a biometric detection system, in accordance with the present invention, provides an edited database including a plurality of existing biometric images with corrected extracted features that were acquired with a sensor or sensors different from the sensor of the system under evaluation. A second database, smaller than the edited database, is edited which includes biometric images (that were acquired by the sensor of the system under evaluation) are employed to evaluate the biometric detection system. The second database has errors in extracted features corrected. A statistical perturbation model is constructed to describe degradation characteristics of the extracted features from the second database as provided in the editing step. The statistical perturbation model is applied to the edited database to construct a perturbed database sensitive to degradations of the biometric system under evaluation. The biometric system is evaluated against the edited database and the perturbed database to predict a performance of the biometric system.
Another method, in accordance with the present invention, for evaluating a biometric detection system, includes the steps of providing a first database having a plurality of biometric images representative of a predetermined population, editing the first database to construct an edited database, the edited database having errors in extracted features from the first database corrected, providing a second database, smaller than the first database, which includes biometric images employed to evaluate the biometric detection system, the second database being representative of a sample of the predetermined population, editing the second database to construct an edited database having errors in extracted features from the second database corrected, constructing a statistical perturbation model to describe degradation characteristics of the extracted features from the second database, applying the statistical perturbation model to the first database to construct a perturbed database sensitive to degradations of the biometric detection system and evaluating the biometric detection system against the first database and the perturbed database to predict a performance of the biometric detection system.
In other systems and methods, the biometric detection system preferably includes a fingerprint recognition system, and the biometric images include images of fingerprints. The step of constructing a statistical perturbation model to describe degradation characteristics of the extracted features from the second database as provided in the editing step may include estimating a difference between an ideal feature extractor and a degraded feature extractor which provided the degradation characteristics. The step of providing an edi
Bromba Manfred
Qian Jianzhong
Wu Zhenyu
Hoff Marc S.
Paschburg Donald B.
Siemens Corporate Research Inc.
Tsai Carol S
LandOfFree
Evaluation system and method for fingerprint verification does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Evaluation system and method for fingerprint verification, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Evaluation system and method for fingerprint verification will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3049024