Multimodal fusion decision logic system using copula model

[0001] This application is a continuation-in-part of and claims the benefit of U.S. patent application Ser. No. 11/273,824 (filed Nov. 15, 2005 now U.S. Pat. No. 7,287,013), which claims the benefit of priority to U.S. provisional patent application Ser. No. 60/643,853 (filed on Jan. 14, 2005). This application also claims the benefit of U.S. provisional patent application Ser. No. 60/970,791 (filed on Sep. 7, 2007).

[0002] The present invention relates to the use of multiple biometric modalities and multiple biometric matching methods in a biometric identification system. Biometric modalities may include (but are not limited to) such methods as fingerprint identification, iris recognition, voice recognition, facial recognition, hand geometry, signature recognition, signature gait recognition, vascular patterns, lip shape, ear shape and palm print recognition.

[0003] Algorithms may be used to combine information from two or more biometric modalities. The combined information may allow for more reliable and more accurate identification of an individual than is possible with systems based on a single biometric modality. The combination of information from more than one biometric modality is sometimes referred to herein as “biometric fusion”.

[0004] Reliable personal authentication is becoming increasingly important. The ability to accurately and quickly identify an individual is important to immigration, law enforcement, computer use, and financial transactions. Traditional security measures rely on knowledge-based approaches, such as passwords and personal identification numbers (“PINs”), or on token-based approaches, such as swipe cards and photo identification, to establish the identity of an individual. Despite being widely used, these are not very secure forms of identification. For example, it is estimated that hundreds of millions of dollars are lost annually in credit card fraud in the United States due to consumer misidentification.

[0005] Biometrics offers a reliable alternative for identifying an individual. Biometrics is the method of identifying an individual based on his or her physiological and behavioral characteristics. Common biometric modalities include fingerprint, face recognition, hand geometry, voice, iris and signature verification. The Federal government will be a leading consumer of biometric applications deployed primarily for immigration, airport, border, and homeland security. Wide scale deployments of biometric applications such as the US-VISIT program are already being done in the United States and else where in the world.

[0006] Despite advances in biometric identification systems, several obstacles have hindered their deployment. Every biometric modality has some users who have illegible biometrics. For example a recent NIST (National Institute of Standards and Technology) study indicates that nearly 2 to 5% of the population does not have legible fingerprints. Such users would be rejected by a biometric fingerprint identification system during enrollment and verification. Handling such exceptions is time consuming and costly, especially in high volume scenarios such as an airport. Using multiple biometrics to authenticate an individual may alleviate this problem.

[0007] Furthermore, unlike password or PIN based systems, biometric systems inherently yield probabilistic results and are therefore not fully accurate. In effect, a certain percentage of the genuine users will be rejected (false non-match) and a certain percentage of impostors will be accepted (false match) by existing biometric systems. High security applications require very low probability of false matches. For example, while authenticating immigrants and international passengers at airports, even a few false acceptances can pose a severe breach of national security. On the other hand false non matches lead to user inconvenience and congestion.

[0008] Existing systems achieve low false acceptance probabilities (also known as False Acceptance Rate or “FAR”) only at the expense of higher false non-matching probabilities (also known as False-Rejection-Rate or “FRR”). It has been shown that multiple modalities can reduce FAR and FRR simultaneously. Furthermore, threats to biometric systems such as replay attacks, spoofing and other subversive methods are difficult to achieve simultaneously for multiple biometrics, thereby making multimodal biometric systems more secure than single modal biometric systems.

[0009] Systematic research in the area of combining biometric modalities is nascent and sparse. Over the years there have been many attempts at combining modalities and many methods have been investigated, including “Logical And”, “Logical Or”, “Product Rule”, “Sum Rule”, “Max Rule”, “Min Rule”, “Median Rule”, “Majority Vote”, “Bayes' Decision”, and “Neyman-Pearson Test”. None of these methods has proved to provide low FAR and FRR that is needed for modern security applications.

[0010] The need to address the challenges posed by applications using large biometric databases is urgent. The US-VISIT program uses biometric systems to enforce border and homeland security. Governments around the world are adopting biometric authentication to implement National identification and voter registration systems. The Federal Bureau of Investigation maintains national criminal and civilian biometric databases for law enforcement.

[0011] Although large-scale databases are increasingly being used, the research community's focus is on the accuracy of small databases, while neglecting the scalability and speed issues important to large database applications. Each of the example applications mentioned above require databases with a potential size in the tens of millions of biometric records. In such applications, response time, search and retrieval efficiency also become important in addition to accuracy.

[0012] The present invention includes a method of deciding whether a data set is acceptable for making a decision. For example, the present invention may be used to determine whether a set of biometrics is acceptable for making a decision about whether a person should be allowed access to a facility. The data set may be comprised of information pieces about objects, such as people. Each object may have at least two types of information pieces, that is to say the data set may have at least two modalities. For example, each object represented in the database may by represented by two or more biometric samples, for example, a fingerprint sample and an iris scan sample. A first probability partition array (“Pm(i,j)”) may be provided. The Pm(i,j) may be comprised of probability values for information pieces in the data set, each probability value in the Pm(i,j) corresponding to the probability of an authentic match. Pm(i,j) may be similar to a Neyman-Pearson Lemma probability partition array. A second probability partition array (“Pfm(i,j)”) may be provided, the Pfm(i,j) being comprised of probability values for information pieces in the data set, each probability value in the Pfm(i,j) corresponding to the probability of a false match. Pfm(i,j) may be similar to a Neyman-Pearson Lemma probability partition array. The Pm(i,j) and/or the Pfm(i,j) may be Copula models.

[0013] A method according to the invention may identify a no-match zone. For example, the no-match zone may be identified by identifying a first index set (“A”), the indices in set A being the (i,j) indices that have values in both Pfm(i,j) and Pm(i,j). A second index set (“Z∞”) may be identified, the indices of Z∞ being the (i,j) indices in set A where both Pfm(i,j) is larger than zero and Pm(i,j) is equal to zero. FAR_Z∞ may be determined, where FAR_Z∞=1−Σ_(i,j)∈Z∞P_fm(i,j). FAR_Z∞ may be compared to a desired false-acceptance-rate (“FAR”), and if FAR_Z∞ is greater than the desired false-acceptance-rate, than the data set may be rejected for failing to provide an acceptable false-acceptance-rate. If FAR_Z∞ is less than or equal to the desired false-acceptance-rate, then the data set may be accepted, if false-rejection-rate is not important.

[0014] If false-rejection-rate is important, further steps may be executed to determine whether the data set should be rejected. The method may further include identifying a third index set ZM∞, the indices of ZM∞ being the (i,j) indices in Z∞ plus those indices where both Pfm(i,j) and Pm(i,j) are equal to zero. A fourth index set (“C”) may be identified, the indices of C being the (i,j) indices that are in A but not ZM∞. The indices of C may be arranged such that

[0015] $\frac{{P_{\mathrm{fm}}\left(i,j\right)}_k}{{P_m\left(i,j\right)}_k}>=\frac{{P_{\mathrm{fm}}\left(i,j\right)}_{k+1}}{{P_m\left(i,j\right)}_{k+1}}$
to provide an arranged C index. A fifth index set (“Cn”) may be identified. The indices of Cn may be the first N (i,j) indices of the arranged C index, where N is a number for which the following is true: FAR_Z∞∪CN=1−Σ_(i,j)∈Z∞P_fm(i,j)−Σ_(i,j)∈CNP_fm(i,j)≦FAR. The FRR may be determined, where FRR=Σ_(i,j)∈CNP_m(i,j), and compared to a desired false-rejection-rate. If FRR is greater than the desired false-rejection-rate, then the data set may be rejected, even though FAR_Z∞ is less than or equal to the desired false-acceptance-rate. Otherwise, the data set may be accepted.

[0016] In another method according to the invention, the false-rejection-rate calculations and comparisons may be executed before the false-acceptance-rate calculations and comparisons. In such a method, a first index set (“A”) may be identified, the indices in A being the (i,j) indices that have values in both Pfm(i,j) and Pm(i,j). A second index set (“Z∞”) may be identified, the indices of Z∞ being the (i,j) indices of A where Pm(i,j) is equal to zero. A third index set (“C”) may be identified, the indices of C being the (i,j) indices that are in A but not Z∞. The indices of C may be arranged such that

[0017] $\frac{{P_{\mathrm{fm}}\left(i,j\right)}_k}{{P_m\left(i,j\right)}_k}>=\frac{{P_{\mathrm{fm}}\left(i,j\right)}_{k+1}}{{P_m\left(i,j\right)}_{k+1}}$
to provide an arranged C index, and a fourth index set (“Cn”) may be identified. The indices of Cn may be the first N (i,j) indices of the arranged C index, where N is a number for which the following is true: FAR_Z∞∪CN=1−Σ_(i,j)∈Z∞P_fm(i,j)−Σ_(i,j)∈CNP_fm(i,j)≦FAR. The FRR may be determined, where FRR=Σ_(i,j)∈CNP_m(i,j), and compared to a desired false-rejection-rate. If the FRR is greater than the desired false-rejection-rate, then the data set may be rejected. If the FRR is less than or equal to the desired false-rejection-rate, then the data set may be accepted, if false-acceptance-rate is not important. If false-acceptance-rate is important, then the FAR_Z∞ may be determined, where FAR_Z∞=1−Σ_(i,j)∈Z∞P_fm(i,j). The FAR_Z∞ may be compared to a desired false-acceptance-rate, and if FAR_Z∞ is greater than the desired false-acceptance-rate, then the data set may be rejected even though FRR is less than or equal to the desired false-rejection-rate. Otherwise, the data set may be accepted.

[0018] The invention may also be embodied as a computer readable memory device for executing any of the methods described above.

[0053] Preliminary Considerations: To facilitate discussion of the invention it may be beneficial to establish some terminology and a mathematical framework. Biometric fusion may be viewed as an endeavor in statistical decision theory [1] [2]; namely, the testing of simple hypotheses. This disclosure uses the term simple hypothesis as used in standard statistical theory: the parameter of the hypothesis is stated exactly. The hypotheses that a biometric score “is authentic” or “is from an impostor” are both simple hypotheses.

[0054] In a match attempt, the N-scores reported from N<∞ biometrics is called the test observation. The test observation, x∈S, where S⊂R^N (Euclidean N-space) is the score sample space, is a vector function of a random variable X from which is observed a random sample (X₁, X₂, . . . , X_N). The distribution of scores is provided by the joint class-conditional probability density function (“pdf”):
f(x|θ)=f(x₁, x₂, . . . , x_N|θ)  (1)
where θ equals either Au to denote the distribution of authentic scores or Im to denote the distribution of impostor scores. So, if θ=Au, Equation (1) is the pdf for authentic distribution of scores and if θ=Im it is the pdf for the impostor distribution. It is assumed that f(x|θ)=f(x₁, x₂, . . . , x_N|θ)>0 on S.

[0055] Given a test observation, the following two simple statistical hypotheses are observed: The null hypothesis, H₀, states that a test observation is an impostor; the alternate hypothesis, H₁, states that a test observation is an authentic match. Because there are only two choices, the fusion logic will partition S into two disjoint sets: R_Au⊂S and R_Im⊂S, where R_Au∩R_Im=Ø and R_Au∪R_Im=S. Denoting the compliment of a set A by A^C, so that R_Au^C=R_Im and R_Im^C=R_Au.

[0056] The decision logic is to accept H₁ (declare a match) if a test observation, x, belongs to R_Au or to accept H₀ (declare no match and hence reject H₁) if x belongs to R_Im. Each hypothesis is associated with an error type: A Type I error occurs when H₀ is rejected (accept H₁) when H₀ is true; this is a false accept (FA); and a Type II error occurs when H₁ is rejected (accept H₀) when H₁ is true; this is a false reject (FR). Thus denoted, respectively, the probability of making a Type I or a Type II error follows:

[0057] $\begin{array}{cc}\mathrm{FAR}=P_{\mathrm{Type}I}=P\left(x\in R_{\mathrm{Au}}|\mathrm{Im}\right)={\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Im}\right)dx& \left(2\right)\\ \mathrm{FRR}=P_{\mathrm{Type}\mathrm{II}}=P\left(x\in R_{\mathrm{Im}}|\mathrm{Au}\right)={\int }_{R_{\mathrm{Im}}}f\left(x|\mathrm{Au}\right)dx& \left(3\right)\end{array}$
So, to compute the false-acceptance-rate (FAR), the impostor score's pdf is integrated over the region which would declare a match (Equation 2). The false-rejection-rate (FRR) is computed by integrating the pdf for the authentic scores over the region in which an impostor (Equation 3) is declared. The correct-acceptance-rate (CAR) is 1-FRR.

[0058] The class-conditional pdf for each individual biometric, the marginal pdf, is assumed to have finite support; that is, the match scores produced by the i^th biometric belong to a closed interval of the real line, γ_i=[α_i, ω_i], where −∞<α_i<ω_i<+∞. Two observations are made. First, the sample space, S, is the Cartesian product of these intervals, i.e., S=γ₁×γ₂× . . . ×γ_N. Secondly, the marginal pdf, which we will often reference, for any of the individual biometrics can be written in terms of the joint pdf:

[0059] $\begin{array}{cc}{f}_i\left(x_i|\theta \right)={\int }_{{\gamma }_j}{\int }_{\forall j\ne i}\dots \int f\left(x|\theta \right)dx& \left(4\right)\end{array}$
For the purposes of simplicity the following definitions are stated:

• • [0060] Definition 1: In a test of simple hypothesis, the correct-acceptance-rate, CAR=1−FRR is known as the power of the test.
  • [0061] Definition 2: For a fixed FAR, the test of simple H₀ versus simple H₁ that has the largest CAR is called most powerful.
  • [0062] Definition 3: Verification is a one-to-one template comparison to authenticate a person's identity (the person claims their identity).
  • [0063] Definition 4: Identification is a one-to-many template comparison to recognize an individual (identification attempts to establish a person's identity without the person having to claim an identity).
    The receiver operation characteristics (ROC) curve is a plot of CAR versus FAR, an example of which is shown in FIG. 3. Because the ROC shows performance for all possible specifications of FAR, as can be seen in the FIG. 3, it is an excellent tool for comparing performance of competing systems, and we use it throughout our analyses.

[0064] Fusion and Decision Methods: Because different applications have different requirements for error rates, there is an interest in having a fusion scheme that has the flexibility to allow for the specification of a Type-I error yet have a theoretical basis for providing the most powerful test, as defined in Definition 2. Furthermore, it would be beneficial if the fusion scheme could handle the general problem, so there are no restrictions on the underlying statistics. That is to say that:

• • [0065] Biometrics may or may not be statistically independent of each other.
  • [0066] The class conditional pdf can be multi-modal (have local maximums) or have no modes.
  • [0067] The underlying pdf is assumed to be nonparametric (no set of parameters define its shape, such as the mean and standard deviation for a Gaussian distribution).

[0068] A number of combination strategies were examined, all of which are listed by Jain [4] on page 243. Most of the schemes, such as “Demptster-Shafer” and “fuzzy integrals” involve training. These were rejected mostly because of the difficulty in analyzing and controlling their performance mechanisms in order to obtain optimal performance. Additionally, there was an interest in not basing the fusion logic on empirical results. Strategies such as SUM, MEAN, MEDIAN, PRODUCT, MIN, MAX were likewise rejected because they assume independence between biometric features.

[0069] When combining two biometrics using a Boolean “AND” or “OR” it is easily shown to be suboptimal when the decision to accept or reject H₁ is based on fixed score thresholds. However, the “AND” and the “OR” are the basic building blocks of verification (OR) and identification (AND). Since we are focused on fixed score thresholds, making an optimal decision to accept or reject H₁ depends on the structure of: R_Au⊂S and R_Im⊂S, and simple thresholds almost always form suboptimal partitions.

[0070] If one accepts that accuracy dominates the decision making process and cost dominates the combination strategy, certain conclusions may be drawn. Consider a two-biometric verification system for which a fixed FAR has been specified. In that system, a person's identity is authenticated if H₁ is accepted by the first biometric OR if accepted by the second biometric. If H₁ is rejected by the first biometric AND the second biometric, then manual intervention is required. If a cost is associated with each stage of the verification process, a cost function can be formulated. It is a reasonable assumption to assume that the fewer people that filter down from the first biometric sensor to the second to the manual check, the cheaper the system.

[0071] Suppose a fixed threshold is used as the decision making process to accept or reject H₁ at each sensor and solve for thresholds that minimize the cost function. It will obtain settings that minimize cost, but it will not necessarily be the cheapest cost. Given a more accurate way to make decisions, the cost will drop. And if the decision method guarantees the most powerful test at each decision point it will have the optimal cost.

[0072] It is clear to us that the decision making process is crucial. A survey of methods based on statistical decision theory reveals many powerful tests such as the Maximum-Likelihood Test and Bayes' Test. Each requires the class conditional probability density function, as given by Equation 1. Some, such as the Bayes' Test, also require the α priori probabilities P(H₀) and P(H₁), which are the frequencies at which we would expect an impostor and authentic match attempt. Generally, these tests do not allow the flexibility of specifying a FAR—they minimize making a classification error.

[0073] In their seminal paper of 1933 [3], Neyman and Pearson presented a lemma that guarantees the most powerful test for a fixed FAR requiring only the joint class conditional pdf for H₀ and H₁. This test may be used as the centerpiece of the biometric fusion logic employed in the invention. The Neyman-Pearson Lemma guarantees the validity of the test. The proof of the Lemma is slightly different than those found in other sources, but the reason for presenting it is because it is immediately amenable to proving the Corollary to the Neyman Pearson Lemma. The corollary states that fusing two biometric scores with Neyman-Pearson, always provides a more powerful test than either of the component biometrics by themselves. The corollary is extended to state that fusing N biometric scores is better than fusing N−1 scores

[0074] Deriving the Neyman-Pearson Test: Let FAR=α be fixed. It is desired to find R_Au⊂S such that

[0075] $\alpha ={\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Im}\right)dx\mathrm{and}{\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Au}\right)dx$
is most powerful. To do this the objective set function is formed that is analogous to Lagrange's Method:

[0076] $u\left(R_{\mathrm{Au}}\right)={\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Au}\right)dx-\lambda \left[{\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Im}\right)dx-\alpha \right]$
where λ≧0 is an undetermined Lagrange multiplier and the external value of u is subject to the constraint

[0077] $\alpha ={\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Im}\right)dx.$
Rewriting the above equation as

[0078] $u\left(R_{\mathrm{Au}}\right)={\int }_{R_{\mathrm{Au}}}\left[f\left(x|\mathrm{Au}\right)-f\left(x|\mathrm{Im}\right)\right]dx+\lambda \alpha$
which ensures that the integrand in the above equation is positive for all x∈R_Au. Let λ≧0, and, recalling that the class conditional pdf is positive on S is assumed, define

[0079] $R_{\mathrm{Au}}=\left\{x:\left[f\left(x|\mathrm{Au}\right)-\lambda f\left(x|\mathrm{Im}\right)\right]>0\right\}=\left\{x:\frac{f\left(x|\mathrm{Au}\right)}{f\left(x|\mathrm{Im}\right)}>\lambda \right\}$
Then u is a maximum if λ is chosen such that

[0080] $\alpha ={\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Im}\right)dx$
is satisfied.

[0081] The Neyman-Pearson Lemma: Proofs of the Neyman-Pearson Lemma can be found in their paper [4] or in many texts [1] [2]. The proof presented here is somewhat different. An “in-common” region, R_IC is established between two partitions that have the same FAR. It is possible this region is empty. Having R_IC makes it easier to prove the corollary presented.

[0082] Neyman-Pearson Lemma: Given the joint class conditional probability density functions for a system of order N in making a decision with a specified FAR=α, let λ be a positive real number, and let

[0083] $\begin{array}{cc}{R}_{\mathrm{Au}}=\left\{x\in S|\frac{f\left(x|\mathrm{Au}\right)}{f\left(x|\mathrm{Im}\right)}>\lambda \right\}& \left(5\right)\\ R_{\mathrm{Au}}^C=R_{\mathrm{Im}}=\left\{x\in S|\frac{f\left(x|\mathrm{Au}\right)}{f\left(x|\mathrm{Im}\right)}\le \lambda \right\}& \left(6\right)\end{array}$
such that

[0084] $\begin{array}{cc}\alpha ={\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Im}\right)dx.& \left(7\right)\end{array}$
then R_Au is the best critical region for declaring a match—it gives the largest correct-acceptance-rate (CAR), hence R_λ^C gives the smallest FRR.

[0085] Proof: The lemma is trivially true if R_Au is the only region for which Equation (7) holds. Suppose R_φ≠R_Au with m(R_φ∩R_Au^C)>0 (this excludes sets that are the same as R_Au except on a set of measure zero that contribute nothing to the integration), is any other region such that

[0086] $\begin{array}{cc}\alpha ={\int }_{R_{\varphi }}f\left(x|\mathrm{Im}\right)dx.& \left(8\right)\end{array}$
Let R_IC=R_φ∩R_Au, which is the “in common” region of the two sets and may be empty. The following is observed:

[0087] $\begin{array}{cc}{\int }_{R_{\mathrm{Au}}}f\left(x|\theta \right)dx={\int }_{R_{\mathrm{Au}}-R_{\mathrm{IC}}}f\left(x|\theta \right)dx+{\int }_{R_{\mathrm{IC}}}f\left(x|\theta \right)dx& \left(9\right)\\ {\int }_{R_{\varphi }}f\left(x|\theta \right)dx={\int }_{R_{\varphi }-R_{\mathrm{IC}}}f\left(x|\theta \right)dx+{\int }_{R_{\mathrm{IC}}}f\left(x|\theta \right)dx& \left(10\right)\end{array}$
If R_Au has a better CAR than R_φ then it is sufficient to prove that

[0088] $\begin{array}{cc}{\int }_{R_{\mathrm{Au}}}f\left(x|\mathrm{Au}\right)dx-{\int }_{R_{\varphi }}f\left(x|\mathrm{Au}\right)dx>0& \left(11\right)\end{array}$
From Equations (7), (8), (9), and (10) it is seen that (11) holds if

[0089] $\begin{array}{cc}{\int }_{R_{\mathrm{Au}}-R_{\mathrm{IC}}}f\left(x|\mathrm{Au}\right)dx-{\int }_{R_{\varphi }-R_{\mathrm{IC}}}f\left(x|\mathrm{Au}\right)dx>0& \left(12\right)\end{array}$
Equations (7), (8), (9), and (10) also gives

[0090] $\begin{array}{cc}{\alpha }_{R_{\varphi }-R_{\mathrm{IC}}}={\int }_{R_{\mathrm{Au}}-R_{\mathrm{IC}}}f\left(x|\mathrm{Im}\right)dx={\int }_{R_{\varphi }-R_{\mathrm{IC}}}f\left(x|\mathrm{Im}\right)dx& \left(13\right)\end{array}$
When x∈R_Au−R_IC⊂R_Au it is observed from (5) that

[0091] $\begin{array}{cc}{\int }_{R_{\mathrm{Au}}-R_{\mathrm{IC}}}f\left(x|\mathrm{Au}\right)dx>{\int }_{R_{\mathrm{Au}}-R_{\mathrm{IC}}}\lambda f\left(x|\mathrm{Im}\right)dx=\lambda {\alpha }_{R_{\varphi }-R_{\mathrm{IC}}}& \left(14\right)\end{array}$
and when x∈R_φ−R_IC⊂R_Au^C it is observed from (6) that

[0092] $\begin{array}{cc}{\int }_{R_{\varphi }-R_{\mathrm{IC}}}f\left(x|\mathrm{Au}\right)dx\le {\int }_{R_{\varphi }-R_{\mathrm{IC}}}\lambda f\left(x|\mathrm{Im}\right)dx=\lambda {\alpha }_{R_{\varphi }-R_{\mathrm{IC}}}& \left(15\right)\end{array}$
Equations (14) and (15) give

[0093] $\begin{array}{cc}{\int }_{R_{\mathrm{Au}}-R_{\mathrm{IC}}}f\left(x|\mathrm{Au}\right)dx>\lambda {\alpha }_{R_{\varphi }-R_{\mathrm{IC}}}\ge {\int }_{R_{\varphi }-R_{\mathrm{IC}}}f\left(x|\mathrm{Au}\right)dx& \left(16\right)\end{array}$
This establishes (12) and hence (11), which proves the lemma. // In this disclosure, the end of a proof is indicated by double slashes “//.”

[0094] Corollary: Neyman-Pearson Test Accuracy Improves with Additional Biometrics: The fact that accuracy improves with additional biometrics is an extremely important result of the Neyman-Pearson Lemma. Under the assumed class conditional densities for each biometric, the Neyman-Pearson Test provides the most powerful test over any other test that considers less than all the biometrics available. Even if a biometric has relatively poor performance and is highly correlated to one or more of the other biometrics, the fused CAR is optimal. The corollary for N-biometrics versus a single component biometric follows.

[0095] Corollary to the Neyman Pearson Lemma: Given the joint class conditional probability density functions for an N-biometric system, choose α=FAR and use the Neyman-Pearson Test to find the critical region R_Au that gives the most powerful test for the N-biometric system

[0096] $\begin{array}{cc}{\mathrm{CAR}}_{R_{\mathrm{Au}}}={\int }_{R_{\mathrm{Au}}}f\left(x❘\mathrm{Au}\right)dx.& \left(17\right)\end{array}$
Consider the i^th biometric. For the same α=FAR, use the Neyman-Pearson Test to find the critical collection of disjoint intervals I_Au⊂R¹ that gives the most powerful test for the single biometric, that is,

[0097] $\begin{array}{cc}{\mathrm{CAR}}_{I_{\mathrm{Au}}}={\int }_{\mathrm{IAu}}f\left(x_i❘\mathrm{Au}\right)dx& \left(18\right)\end{array}$
then
CAR_RAu≧CAR_IAu.  (19)
Proof: Let R_i=I_Au×γ₁×γ₂× . . . ×γ_N⊂R^N, where the Cartesian products are taken over all the γ_k except for k=i. From (4) the marginal pdf can be recast in terms of the joint pdf

[0098] $\begin{array}{cc}{\int }_{R_i}f\left(x❘\theta \right)dx={\int }_{I_{\mathrm{Au}}}f\left(x_i❘\theta \right)dx& \left(20\right)\end{array}$
First, it will be shown that equality holds in (19) if and only if R_Au=R_i. Given that R_Au=R_i except on a set of measure zero, i.e., m(R_Au^C∩R_i)=0, then clearly CAR_RAu=CAR_IAu. On the other hand, assume CAR_RAu=CAR_IAu and m(R_Au^C∩R_i)>0, that is, the two sets are measurably different. But this is exactly the same condition previously set forth in the proof of the Neyman-Pearson Lemma from which from which it was concluded that CAR_RAu>CAR_IAu, which is a contradiction. Hence equality holds if and only if R_Au=R_i. Examining the inequality situation in equation (19), given that R_Au≠R_i such that m(R_Au^C∩R_i)>0, then it is shown again that the exact conditions as in the proof of the Neyman-Pearson Lemma have been obtained from which we conclude that CAR_RAu>CAR_IAu, which proves the corollary. //

[0099] Examples have been built such that CAR_RAu=CAR_IAu but it is hard to do and it is unlikely that such densities would be seen in a real world application. Thus, it is safe to assume that CAR_RAu>CAR_IAu is almost always true.

[0100] The corollary can be extended to the general case. The fusion of N biometrics using Neyman-Pearson theory always results in a test that is as powerful as or more powerful than a test that uses any combination of M<N biometrics. Without any loss to generality, arrange the labels so that the first M biometrics are the ones used in the M<N fusion. Choose α=FAR and use the Neyman-Pearson Test to find the critical region R_M⊂R^M that gives the most powerful test for the M-biometric system. Let R_N=R_M×γ_M+1×γ_M+2× . . . ×γ_N⊂R^N, where the Cartesian products are taken over all the γ-intervals not used in the M biometrics combination. Then writing

[0101] $\begin{array}{cc}{\int }_{R_N}f\left(x❘\theta \right)dx={\int }_{R_M}f\left(x_1,x_2\dots ,x_M❘\theta \right)dx& \left(21\right)\end{array}$
gives the same construction as in (20) and the proof flows as it did for the corollary.

[0102] We now state and prove the following five mathematical propositions. The first four propositions are necessary to proving the fifth proposition, which will be cited in the section that details the invention. For each of the propositions we will use the following: Let r={r₁, r₂, . . . , r_n} be a sequence of real valued ratios such that r₁≧r₂≧ . . . ≧r_n. For each r_i we know the numerator and denominator of the ratio, so that

[0103] $r_i=\frac{n_i}{d_i},n_i,d_i>0\forall i.$

[0104] Proposition #1: For r_i,r_i+1∈r and

[0105] $r_i=\frac{n_i}{d_i},r_{i+1}=\frac{n_{i+1}}{d_{i+1}},$
then

[0106] $\frac{n_i+n_{i+1}}{d_i+d_{i+1}}\le \frac{n_i}{d_i}=r_i.$
Proof: Because r_i,r_i+1∈r we have

[0107] $\frac{n_{i+1}}{d_{i+1}}\le \frac{n_i}{d_i}\Rightarrow \frac{d_i{n}_{i+1}}{n_i{d}_{i+1}}\le 1\Rightarrow \frac{d_i}{n_i}\left(n_i+n_{i+1}\right)\le d_i+d_{i+1}\Rightarrow \frac{n_i+n_{i+1}}{d_i+d_{i+1}}\le \frac{n_i}{d_i}=r_i$

• • [0108] //

[0109] Proposition #2: For

[0110] $r_i,r_{i+1}\in r\mathrm{and}{r}_i=\frac{n_i}{d_i},r_{i+1}=\frac{n_{i+1}}{d_{i+1}},\mathrm{then}$ $r_{i+1}=\frac{n_{i+1}}{d_{i+1}}\le \frac{n_i+n_{i+1}}{d_i+d_{i+1}}.$
Proof: The proof is the same as for proposition #1 except the order is reversed.

[0111] Proposition #3:

[0112] $\mathrm{For}{r}_i=\frac{n_i}{d_i},i=\left(1,\dots ,n\right),\mathrm{then}\frac{\sum _{i=m}^M{n}_i}{\sum _{i=m}^M{d}_i}\le \frac{n_m}{d_m},1\le m\le M\le n.$
Proof: The proof is by induction. We know from proposition #1 that the result holds for N=1, that is

[0113] $\frac{n_i+n_{i+1}}{d_i+d_{i+1}}\le \frac{n_i}{d_i}.$
Now assume it holds for any N>1, we need to show that it holds for N+1. Let m=2 and M=N+1 (note that this is the N^th case and not the N+1 case), then we assume

[0114] $\frac{\sum _{i=2}^{N+1}{n}_i}{\sum _{i=2}^{N+1}{d}_i}\le \frac{n_2}{d_2}\le \frac{n_1}{d_1}\Rightarrow \frac{d_1}{n_1}\frac{\sum _{i=2}^{N+1}{n}_i}{\sum _{i=2}^{N+1}{d}_i}\le 1\Rightarrow \frac{d_1}{n_1}\sum _{i=2}^{N+1}{n}_i\le \sum _{i=2}^{N+1}{d}_i\Rightarrow \frac{d_1}{n_1}\sum _{i=2}^{N+1}{n}_i+d_i\le \sum _{i=2}^{N+1}{d}_i+d_1=\sum _{i=1}^{N+1}{d}_i\Rightarrow d_1\left(1+\frac{\sum _{i=2}^{N+1}{n}_i}{n_1}\right)\le \sum _{i=1}^{N+1}{d}_i\Rightarrow d_1\left(\frac{n_1+\sum _{i=2}^{N+1}{n}_i}{n_1}\right)=\frac{d_1}{n_1}\sum _{i=1}^{N+1}{n}_i\le \sum _{i=1}^{N+1}{d}_i\Rightarrow \frac{\sum _{i=1}^{N+1}{n}_i}{\sum _{i=1}^{N+1}{d}_i}\le \frac{n_1}{d_1}$
which is the N+1 case as was to be shown.

• • [0115] //

[0116] Proposition #4:

[0117] $\mathrm{For}{r}_i=\frac{n_i}{d_i},i=\left(1,\dots ,n\right),\mathrm{then}\frac{n_M}{d_M}\le \frac{\sum _{i=m}^M{n}_i}{\sum _{i=m}^M{d}_i},$
1≦m≦M≦n. Proof: As with proposition #3, the proof is by induction. We know from proposition #2 that the result holds for N=1, that is

[0118] $\frac{n_{i+1}}{d_{i+1}}\le \frac{n_i+n_{i+1}}{d_i+d_{i+1}}.$
The rest of the proof follows the same format as for proposition #3 with the order reversed.

• • [0119] //

[0120] Proposition #5: Recall that the r is an ordered sequence of decreasing ratios with known numerators and denominators. We sum the first N numerators to get S_n and sum the first N denominators to get S_d. We will show that for the value S_n, there is no other collection of ratios in r that gives the same S_n and a smaller S_d. For

[0121] $r_i=\frac{n_i}{d_i},i=\left(1,\dots ,n\right),$
let S be the sequence of the first N terms of r, with the sum of numerators given by

[0122] $S_n=\sum _{i=1}^N{n}_i,$
and the sum of denominators by

[0123] $S_d=\sum _{i=1}^N{d}_i,$
1≦N≦n. Let S′ be any other sequence of ratios in r, with numerator sum S′_n=Σn_i and denominator sum S′_d=Σd_i such that S′_n=S_n, then we have S_d≦S′_d. Proof: The proof is by contradiction. Suppose the proposition is not true, that is, assume another sequence, S′, exists such that S′_d<S_d. For this sequence, define the indexing sets A={indices that can be between 1 and N inclusive} and B={indices that can be between N+1 and n inclusive}. We also define the indexing set A^c={indices between 1 and N inclusive and not in A}, which means A∩A^c=Ø and A∪A^c={all indices between 1 and N inclusive}. Then our assumption states:

[0124] $\begin{array}{c}{S}_n^{{}^{\prime }}=\sum _{i\in A}^{}{n}_i+\sum _{i\in B}^{}{n}_i=S_n\\ \mathrm{and}\\ S_d^{{}^{\prime }}=\sum _{i\in A}^{}{d}_i+\sum _{i\in B}^{}{d}_i<S_d.\end{array}$
This implies that

[0125] $\begin{array}{cc}\sum _{i\in A}^{}{d}_i+\sum _{i\in B}^{}{d}_i<\sum _{i=1}^N{d}_i=\sum _{i\in A}^{}{d}_i+\sum _{i\in A^c}^{}{d}_i.\Rightarrow \sum _{i\in B}^{}{d}_i<\sum _{i\in A^c}^{}{d}_i\Rightarrow \frac{1}{\sum _{i\in A^c}^{}{d}_i}<\frac{1}{\sum _{i\in B}^{}{d}_i}& \left(22\right)\end{array}$
Because the numerators of both sequences are equal, we can write

[0126] $\begin{array}{cc}\begin{array}{c}\sum _{i\in A}^{}{n}_i+\sum _{i\in B}^{}{n}_i=\sum _{i\in A⋓A^c}^{}{n}_i\\ \Rightarrow \sum _{i\in B}^{}{n}_i=\sum _{i\in A⋓A^c}^{}{n}_i-\sum _{i\in A}^{}{n}_i\\ =\sum _{i\in A^c}^{}{n}_i\end{array}& \left(23\right)\end{array}$
Combining (22) and (23), and from propositions #3 and #4, we have

[0127] $r_N=\frac{n_N}{d_N}\le \frac{\sum _{i\in A^c}{n}_i}{\sum _{i\in A^c}{d}_i}<\frac{\sum _{i\in B}{n}_i}{\sum _{i\in B}{d}_i}\le \frac{n_{N+1}}{d_{N+1}}=r_{N+1},$
which contradicts the fact that r_N≧r_N+1, hence the validity of the proposition follows.

• • [0128] //

[0129] Cost Functions for Optimal Verification and Identification: In the discussion above, it is assumed that all N biometric scores are simultaneously available for fusion. The Neyman-Pearson Lemma guarantees that this provides the most powerful test for a fixed FAR. In practice, however, this could be an expensive way of doing business. If it is assumed that all aspects of using a biometric system, time, risk, etc., can be equated to a cost, then a cost function can be constructed. The inventive process described below constructs the cost function and shows how it can be minimized using the Neyman-Pearson test. Hence, the Neyman-Pearson theory is not limited to “all-at-once” fusion; it can be used for serial, parallel, and hierarchal systems.

[0130] Having laid a basis for the invention, a description of two cost functions is now provided as a mechanism for illustrating an embodiment of the invention. A first cost function for a verification system and a second cost function for an identification system are described. For each system, an algorithm is presented that uses the Neyman-Pearson test to minimize the cost function for a second order biometric system, that is a biometric system that has two modalities. The cost functions are presented for the general case of N-biometrics. Because minimizing the cost function is recursive, the computational load grows exponentially with added dimensions. Hence, an efficient algorithm is needed to handle the general case.

[0131] First Cost Function—Verification System. A cost function for a 2-stage biometric verification (one-to-one) system will be described, and then an algorithm for minimizing the cost function will be provided. In a 2-stage verification system, a subject may attempt to be verified by a first biometric device. If the subject's identity cannot be authenticated, the subject may attempt to be authenticated by a second biometric. If that fails, the subject may resort to manual authentication. For example, manual authentication may be carried out by interviewing the subject and determining their identity from other means.

[0132] The cost for attempting to be authenticated using a first biometric, a second biometric, and a manual check are c₁, c₂, and c₃, respectively. The specified FAR_sys is a system false-acceptance-rate, i.e. the rate of falsely authenticating an impostor includes the case of it happening at the first biometric or the second biometric. This implies that the first-biometric station test cannot have a false-acceptance-rate, FAR₁, that exceeds FAR_sys. Given a test with a specified FAR₁, there is an associated false-rejection-rate, FRR₁, which is the fraction of subjects that, on average, are required to move on to the second biometric station. The FAR required at the second station is FAR₂=FAR_sys−FAR₁. It is known that P(A∪B)=P(A)+P(B)−P(A)P(B), so the computation of FAR₂ appears imperfect. However, if FAR_sys=P(A∪B) and FAR₁=P(A), in the construction of the decision space, it is intended that FAR₂=P(B)−P(A)P(B).

[0133] Note that FAR₂ is a function of the specified FAR_sys and the freely chosen FAR₁; FAR₂ is not a free parameter. Given a biometric test with the computed FAR₂, there is an associated false-rejection-rate, FRR₂ by the second biometric test, which is the fraction of subjects that are required to move on to a manual check. This is all captured by the following cost function:
Cost=c₁+FRR₁(c₂+c₃FRR₂)  (30)

[0134] There is a cost for every choice of FAR₁≦FAR_sys, so the Cost in Equation 30 is a function of FAR₁. For a given test method, there exists a value of FAR₁ that yields the smallest cost and we present an algorithm to find that value. In a novel approach to the minimization of Equation 30, a modified version of the Neyman-Pearson decision test has been developed so that the smallest cost is optimally small.

[0135] An algorithm is outlined below. The algorithm seeks to optimally minimize Equation 30. To do so, we (a) set the initial cost estimate to infinity and, (b) for a specified FAR_sys, loop over all possible values of FAR₁≦FAR_sys. In practice, the algorithm may use a uniformly spaced finite sample of the infinite possible values. The algorithm may proceed as follows: (c) set FAR₁=FAR_sys, (d) set Cost=∞, and (e) loop over possible FAR₁ values. For the first biometric at the current FAR₁ value, the algorithm may proceed to (f) find the optimal Match-Zzone, R₁, and (g) compute the correct-acceptance-rate over R₁ by:

[0136] $\begin{array}{cc}{\mathrm{CAR}}_1={\int }_{R_1}f\left(x|\mathrm{Au}\right)dx& \left(31\right)\end{array}$
and (h) determine FRR₁ using FRR₁=1−CAR₁.

[0137] Next, the algorithm may test against the second biometric. Note that the region R₁ of the score space is no longer available since the first biometric test used it up. The Neyman-Pearson test may be applied to the reduced decision space, which is the compliment of R₁. So, at this time, (h) the algorithm may compute FAR₂=FAR_sys−FAR₁, and FAR₂ may be (i) used in the Neyman-Pearson test to determine the most powerful test, CAR₂, for the second biometric fused with the first biometric over the reduced decision space R₁^C. The critical region for CAR₂ is R₂, which is disjoint from R₁ by our construction. Score pairs that result in the failure to be authenticated at either biometric station must fall within the region R₃=(R₁∪R₂)^C, from which it is shown that

[0138] ${\mathrm{FRR}}_2=\frac{{\int }_{R_3}f\left(x|\mathrm{Au}\right)dx}{{\mathrm{FRR}}_1}$

[0139] The final steps in the algorithm are (j) to compute the cost using Equation 30 at the current setting of FAR₁ using FRR₁ and FRR₂, and (k) to reset the minimum cost if cheaper, and keep track of the FAR₁ responsible for the minimum cost.

[0140] To illustrate the algorithm, an example is provided. Problems arising from practical applications are not to be confused with the validity of the Neyman-Pearson theory. Jain states in [5]: “In case of a larger number of classifiers and relatively small training data, a classifier my actually degrade the performance when combined with other classifiers . . . ” This would seem to contradict the corollary and its extension. However, the addition of classifiers does not degrade performance because the underlying statistics are always true and the corollary assumes the underlying statistics. Instead, degradation is a result of inexact estimates of sampled densities. In practice, a user may be forced to construct the decision test from the estimates, and it is errors in the estimates that cause a mismatch between predicted performance and actual performance.

[0141] Given the true underlying class conditional pdf for, H₀ and H₁, the corollary is true. This is demonstrated with a challenging example using up to three biometric sensors. The marginal densities are assumed to be Gaussian distributed. This allows a closed form expression for the densities that easily incorporates correlation. The general n^th order form is well known and is given by

[0142] $\begin{array}{cc}f\left(x|\theta \right)=\frac{1}{{\left(2\pi \right)}^{\frac{n}{2}}{C}^{1/2}}\exp \left(-\frac{1}{2}\left[{\left(x-\mu \right)}^T{C}^{-1}\left(x-\mu \right)\right]\right)& \left(32\right)\end{array}$
where μ is the mean and C is the covariance matrix. The mean (μ) and the standard deviation (σ) for the marginal densities are given in Table 1. Plots of the three impostor and three authentic densities are shown in FIG. 1.

[0143]

TABLE 1
Biometric	Impostors		Authentics
Number	μ	Σ	μ	σ
#1	0.3	0.06	0.65	0.065
#2	0.32	0.065	0.58	0.07
#3	0.45	0.075	0.7	0.08

[0144]

TABLE 2
Correlation Coefficient	Impostors	Authentics
ρ12	0.03	0.83
ρ13	0.02	0.80
ρ23	0.025	0.82

[0145] To stress the system, the authentic distributions for the three biometric systems may be forced to be highly correlated and the impostor distributions to be lightly correlated. The correlation coefficients (ρ) are shown in Table 2. The subscripts denote the connection. A plot of the joint pdf for the fusion of system #1 with system #2 is shown in FIG. 2, where the correlation between the authentic distributions is quite evident.

[0146] The single biometric ROC curves are shown in FIG. 3. As could be predicted from the pdf curves plotted in FIG. 1, System #1 performs much better than the other two systems, with System #3 having the worst performance.

[0147] Fusing 2 systems at a time; there are three possibilities: #1+#2, #1+#3, and #2+#3. The resulting ROC curves are shown in FIG. 4. As predicted by the corollary, each 2-system pair outperforms their individual components. Although the fusion of system #2 with system #3 has worse performance than system #1 alone, it is still better than the single system performance of either system #2 or system #3.

[0148] Finally, FIG. 5 depicts the results when fusing all three systems and comparing its performance with the performance of the 2-system pairs. The addition of the third biometric system gives substantial improvement over the best performing pair of biometrics.

[0149] Tests were conducted on individual and fused biometric systems in order to determine whether the theory presented above accurately predicts what will happen in a real-world situation. The performance of three biometric systems were considered. The numbers of score samples available are listed in Table 3. The scores for each modality were collected independently from essentially disjoint subsets of the general population.

[0150]

	TABLE 3
		Number of	Number of
	Biometric	Impostor Scores	Authentic Scores
	Fingerprint	8,500,000	21,563
	Signature	325,710	990
	Facial Recognition	4,441,659	1,347

[0151] To simulate the situation of an individual obtaining a score from each biometric, the initial thought was to build a “virtual” match from the data of Table 3. Assuming independence between the biometrics, a 3-tuple set of data was constructed. The 3-tuple set was an ordered set of three score values, by arbitrarily assigning a fingerprint score and a facial recognition score to each signature score for a total of 990 authentic score 3-tuples and 325,710 impostor score 3-tuples.

[0152] By assuming independence, it is well known that the joint pdf is the product of the marginal density functions, hence the joint class conditional pdf for the three biometric systems, f(x|θ)=f(x₁, x₂, x₃|θ) can be written as
f(x|θ)=f_fingerprint(x|θ)f_signature(x|θ)f_facial(x|θ)  (33)

[0153] So it is not necessary to dilute the available data. It is sufficient to approximate the appropriate marginal density functions for each modality using all the data available, and compute the joint pdf using Equation 33.

[0154] The class condition density functions for each of the three modalities, f_fingerprint(x|θ), f_signature(x|θ) and f_facial(x|θ), were estimated using the available sampled data. The authentic score densities were approximated using two methods: (1) the histogram-interpolation technique and (2) the Parzen-window method. The impostor densities were approximated using the histogram-interpolation method. Although each of these estimation methods are guaranteed to converge in probability to the true underlying density as the number of samples goes to infinity, they are still approximations and can introduce error into the decision making process, as will be seen in the next section. The densities for fingerprint, signature, and facial recognition are shown in FIG. 6, FIG. 7 and FIG. 8 respectively. Note that the authentic pdf for facial recognition is bimodal. The Neyman-Pearson test was used to determine the optimal ROC for each of the modalities. The ROC curves for fingerprint, signature, and facial recognition are plotted in FIG. 9.

[0155] There are three possible unique pairings of the three biometric systems: (1) fingerprints with signature, (2) fingerprints with facial recognition, and (3) signatures with facial recognition. Using the marginal densities (above) to create the required 2-D joint class conditional density functions, two sets of 2-D joint density functions were computed—one in which the authentic marginal densities were approximated using the histogram method, and one in which the densities were approximated using the Parzen window method. Using the Neyman-Pearson test, an optimal ROC was computed for each fused pairing and each approximation method. The ROC curves for the histogram method are shown in FIG. 10 and the ROC curves for the Parzen window method are shown in FIG. 11.

[0156] As predicted by the corollary, the fused performance is better than the individual performance for each pair under each approximation method. But, as we cautioned in the example, error due to small sample sizes can cause pdf distortion. This is apparent when fusing fingerprints with signature data (see FIG. 10 and FIG. 11). Notice that the Parzen-window ROC curve (FIG. 11) crosses over the curve for fingerprint-facial-recognition fusion, but does not cross over when using the histogram interpolation method (FIG. 10). Small differences between the two sets of marginal densities are magnified when using their product to compute the 2-dimensional joint densities, which is reflected in the ROC.

[0157] As a final step, all three modalities were fused. The resulting ROC using histogram interpolating is shown in FIG. 12, and the ROC using the Parzen window is shown FIG. 13. As might be expected, the Parzen window pdf distortion with the 2-dimensional fingerprint-signature case has carried through to the 3-dimensional case. The overall performance, however, is dramatically better than any of the 2-dimensional configurations as predicted by the corollary.

[0158] In the material presented above, it was assumed that all N biometric scores would be available for fusion. Indeed, the Neyman-Pearson Lemma guarantees that this provides the most powerful test for a fixed FAR. In practice, however, this could be an expensive way of doing business. If it is assumed that all aspects of using a biometric system, time, risk, etc., can be equated to a cost, then a cost function can be constructed. Below, we construct the cost function and show how it can be minimized using the Neyman-Pearson test. Hence, the Neyman-Pearson theory is not limited to “all-at-once” fusion—it can be used for serial, parallel, and hierarchal systems.

[0159] In the following section, a review of the development of the cost function for a verification system is provided, and then a cost function for an identification system is developed. For each system, an algorithm is presented that uses the Neyman-Pearson test to minimize the cost function for a second order modality biometric system. The cost function is presented for the general case of N-biometrics. Because minimizing the cost function is recursive, the computational load grows exponentially with added dimensions. Hence, an efficient algorithm is needed to handle the general case.

[0160] From the prior discussion of the cost function for a 2-station system, the costs for using the first biometric, the second biometric, and the manual check are c₁, c₂, and c₃, respectively, and FAR_sys is specified. The first-station test cannot have a false-acceptance-rate, FAR₁, that exceeds FAR_sys. Given a test with a specified FAR₁, there is an associated false-rejection-rate, FRR₁, which is the fraction of subjects that, on average, are required to move on to the second station. The FAR required at the second station is FAR₂=FAR_sys−FAR₁. It is known that P(A∪B)=P(A)+P(B)−P(A)P(B), so the computation of FAR₂ appears imperfect. However, if FAR_sys=P(A∪B) and FAR₁=P(A), in the construction of the decision space, it is intended that FAR₂=P(B)−P(A)P(B).

[0161] Given a test with the computed FAR₂, there is an associated false-rejection-rate, FRR₂ by the second station, which is the fraction of subjects that are required to move on to a manual checkout station. This is all captured by the following cost function
Cost=c₁+FRR₁(c₂+c₃FRR₂)  (30)

[0162] There is a cost for every choice of FAR₁≦FAR_sys, so the Cost in Equation 30 is a function of FAR₁. For a given test method, there exists a value of FAR₁ that yields the smallest cost and we develop an algorithm to find that value.

[0163] Using a modified Neyman-Pearson test to optimally minimize Equation 30, an algorithm can be derived. Step 1: set the initial cost estimate to infinity and, for a specified FAR_sys, loop over all possible values of FAR₁≦FAR_sys. To be practical, in the algorithm a finite sample of the infinite possible values may be used. The first step in the loop is to use the Neyman-Pearson test at FAR₁ to determine the most powerful test, CAR₁, for the first biometric, and FRR₁=1−CAR₁ is computed. Since it is one dimensional, the critical region is a collection of disjoint intervals I_Au. As in the proof to the corollary, the 1-dimensional I_Au is recast as a 2-dimensional region, R₁=I_Au×γ₂⊂R², so that

[0164] $\begin{array}{cc}{\mathrm{CAR}}_1={\int }_{R_1}f\left(x|\mathrm{Au}\right)dx={\int }_{1_{\mathrm{Au}}}f\left(x_1|\mathrm{Au}\right)dx& \left(34\right)\end{array}$

[0165] When it is necessary to test against the second biometric, the region R₁ of the decision space is no longer available since the first test used it up. The Neyman-Pearson test can be applied to the reduced decision space, which is the compliment of R₁. Step 2: FAR₂=FAR_sys−FAR₁ is computed. Step 3: FAR₂ is used in the Neyman-Pearson test to determine the most powerful test, CAR₂, for the second biometric fused with the first biometric over the reduced decision space R₁^C. The critical region for CAR₂ is R₂, which is disjoint from R₁ by the construction. Score pairs that result in the failure to be authenticated at either biometric station must fall within the region R₃=(R₁∪R₂)^C, from which it is shown that

[0166] ${\mathrm{FRR}}_2=\frac{{\int }_{R_3}f\left(x|\mathrm{Au}\right)dx}{{\mathrm{FRR}}_1}$

[0167] Step 4: compute the cost at the current setting of FAR₁ using FRR₁ and FRR₂. Step 5: reset the minimum cost if cheaper, and keep track of the FAR₁ responsible for the minimum cost. A typical cost function is shown in FIG. 14.

[0168] Two special cases should be noted. In the first case, if c₁>0, c₂>0, and c₃=0, the algorithm shows that the minimum cost is to use only the first biometric—that is, FAR₁=FAR_sys. This makes sense because there is no cost penalty for authentication failures to bypass the second station and go directly to the manual check.

[0169] In the second case, c₁=c₂=0 and c₃>0, the algorithm shows that scores should be collected from both stations and fused all at once; that is, FAR₁=1.0. Again, this makes sense because there is no cost penalty for collecting scores at both stations, and because the Neyman-Pearson test gives the most powerful CAR (smallest FRR) when it can fuse both scores at once.

[0170] To extend the cost function to higher dimensions, the logic discussed above is simply repeated to arrive at

[0171] $\begin{array}{cc}\mathrm{Cost}=c_1+c_2{\mathrm{FRR}}_1+c_3\prod _{i=1}^2{\mathrm{FRR}}_i+\dots +c_{N+1}\prod _{i=1}^N{\mathrm{FRR}}_i& \left(35\right)\end{array}$
In minimizing Equation 35, there is 1 degree of freedom, namely FAR₁. Equation 35 has N−1 degrees of freedom. When FAR₁≦FAR_sys are set, then FAR₂≦FAR₁ can bet set, then FAR₃≦FAR₂, and so on—and to minimize thus, N−1 levels of recursion.

[0172] Identification System: Identification (one-to-many) systems are discussed. With this construction, each station attempts to discard impostors. Candidates that cannot be discarded are passed on to the next station. Candidates that cannot be discarded by the biometric systems arrive at the manual checkout. The goal, of course, is to prune the number of impostor templates thus limiting the number that move on to subsequent steps. This is a logical AND process—for an authentic match to be accepted, a candidate must pass the test at station 1 and station 2 and station 3 and so forth. In contrast to a verification system, the system administrator must specify a system false-rejection-rate, FRR_sys instead of a FAR_sys. But just like the verification system problem the sum of the component FRR values at each decision point cannot exceed FRR_sys. If FRR with FAR are replaced in Equations 34 or 35 the following cost equations are arrived at for 2-biometric and an N-biometric identification system

[0173] $\begin{array}{cc}\mathrm{Cost}=c_1+{\mathrm{FAR}}_1\left(c_2+c_3{\mathrm{FAR}}_2\right)& \left(36\right)\\ \mathrm{Cost}=c_1+c_2{\mathrm{FAR}}_1+c_3\prod _{i=1}^2{\mathrm{FAR}}_i+\dots +c_{N+1}\prod _{i=1}^N{\mathrm{FAR}}_i& \left(37\right)\end{array}$
Alternately, equation 37 may be written as:

[0174] $\begin{array}{cc}\mathrm{Cost}=\sum _{j=0}^N\left(c_{j+1}\prod _{i=1}^j{\mathrm{FAR}}_i\right),& \left(37A\right)\end{array}$
These equations are a mathematical dual of Equations 34 and 35 are thus minimized using the logic of the algorithm that minimizes the verification system.

[0175] Algorithm: Generating Matching and Non-Matching PDF Surfaces. The optimal fusion algorithm uses the probability of an authentic match and the probability of a false match for each p_ij∈P. These probabilities may be arrived at by numerical integration of the sampled surface of the joint pdf. A sufficient number of samples may be generated to get a “smooth” surface by simulation. Given a sequence of matching score pairs and non-matching score pairs, it is possible to construct a numerical model of the marginal cumulative distribution functions (cdf). The distribution functions may be used to generate pseudo random score pairs. If the marginal densities are independent, then it is straightforward to generate sample score pairs independently from each cdf. If the densities are correlated, we generate the covariance matrix and then use Cholesky factorization to obtain a matrix that transforms independent random deviates into samples that are appropriately correlated.

[0176] Assume a given partition P. The joint pdf for both the authentic and impostor cases may be built by mapping simulated score pairs to the appropriate p_ij∈P and incrementing a counter for that sub-square. That is, it is possible to build a 2-dimensional histogram, which is stored in a 2-dimensional array of appropriate dimensions for the partition. If we divide each array element by the total number of samples, we have an approximation to the probability of a score pair falling within the associated sub-square. We call this type of an array the probability partition array (PPA). Let P_fm be the PPA for the joint false match distribution and let P_m be the PPA for the authentic match distribution. Then, the probability of an impostor's score pair, (s₁, s₂)∈p_ij, resulting in a match is P_fm(i,j). Likewise, the probability of a score pair resulting in a match when it should be a match is P_m(i,j). The PPA for a false reject (does not match when it should) is P_fr=1−P_m.

[0177] Consider the partition arrays, P_fm and P_m, defined in Section #5. Consider the ratio

[0178] $\frac{P_{\mathrm{fm}}\left(i,j\right)}{P_m\left(i,j\right)},P_{\mathrm{fr}}>0.$
The larger this ratio the more the sub-square indexed by (i,j) favors a false match over a false reject. Based on the propositions presented in Section 4, it is optimal to tag the sub-square with the largest ratio as part of the no-match zone, and then tag the next largest, and so on.

[0179] Therefore, an algorithm that is in keeping with the above may proceed as:

• • [0180] Step 1. Assume a required FAR has been given.
  • [0181] Step 2. Allocate the array P_mz to have the same dimensions as P_m and P_fm. This is the match zone array. Initialize all of its elements as belonging to the match zone.
  • [0182] Step 3. Let the index set A={all the (i,j) indices for the probability partition arrays}.
  • [0183] Step 4. Identify all indices in A such that P_m(i,j)=P_fm(i,j)=0 and store those indices in the indexing set Z={(i,j): P_m(i,j)=P_fm(i,j)=0}. Tag each element in P_mz indexed by Z as part of the no-match zone.
  • [0184] Step 5. Let B=A−Z={(i,j): P_m(i,j)≠0 and P_fm(i,j)≠0}. We process only those indices in B. This does not affect optimality because there is most likely zero probability that either a false match score pair or a false reject score pair falls into any sub-square indexed by elements of Z.
  • [0185] Step 6. Identify all indices in B such that P_fm(i,j)>0 and P_m(i,j)=0 and store those indices in Z_∞={(i,j)_k: P_fm(i,j)>0, P_m(i,j)=0}. Simply put, this index set includes the indexes to all the sub-squares that have zero probability of a match but non-zero probability of false match.
  • [0186] Step 7. Tag all the sub-squares in P_mz indexed by Z_∞ as belonging to the no-match zone. At this point, the probability of a matching score pair falling into the no-match zone is zero. The probability of a non-matching score pair falling into the match zone is:
    FAR_Z∞=1−Σ_(i,j)∈Z∞P_fm(i,j)
  • [0187] Furthermore, if FAR_Z∞<=FAR then we are done and can exit the algorithm.
  • [0188] Step 8. Otherwise, we construct a new index set

[0189] $C=A-Z-Z_{\infty }=\left\{{\left(i,j\right)}_k\text{:}{P_{\mathrm{fm}}\left(i,j\right)}_k>0,{P_m\left(i,j\right)}_k>0,\frac{{P_{\mathrm{fm}}\left(i,j\right)}_k}{{P_m\left(i,j\right)}_k}>=\frac{{P_{\mathrm{fm}}\left(i,j\right)}_{k+1}}{{P_m\left(i,j\right)}_{k+1}}\right\}.$

• • [0190] We see that C holds the indices of the non-zero probabilities in a sorted order—the ratios of false-match to match probabilities occur in descending order.
  • [0191] Step 9. Let C_N be the index set that contains the first N indices in C. We determine N so that:
    FAR_Z∞∪CN=1−Σ_(i,j)∈Z∞P_fm(i,j)−Σ_(i,j)∈CNP_fm(i,j)≦FAR
  • [0192] Step 10. Label elements of P_mz indexed by members of C_N as belonging to the no-match zone. This results in a FRR given by
    FRR=Σ_(i,j)∈CNP_m(i,j),
  • [0193]  and furthermore this FRR is optimal.

[0194] It will be recognized that other variations of these steps may be made, and still be within the scope of the invention. For clarity, the notation “(i,j)” is used to identify arrays that have at least two modalities. Therefore, the notation “(i,j)” includes more than two modalities, for example (i,j,k), (i,j,k,l), (i,j,k,l,m), etc.

[0195] For example, FIG. 15 illustrates one method according to the invention in which Pm(i,j) is provided 10 and Pfm(i,j) 13 is provided. As part of identifying 16 indices (i,j) corresponding to a no-match zone, a first index set (“A”) may be identified. The indices in set A may be the (i,j) indices that have values in both Pfm(i,j) and Pm(i,j). A second index set (“Z∞”) may be identified, the indices of Z∞ being the (i,j) indices in set A where both Pfm(i,j) is larger than zero and Pm(i,j) is equal to zero. Then determine FAR_Z∞ 19, where FAR_Z∞=1−Σ_(i,j)∈Z∞P_fm(i,j). It should be noted that the indices of Z∞ may be the indices in set A where Pm(i,j) is equal to zero, since the indices where both Pfm(i,j)=Pm(i,j)=0 will not affect FAR_Z∞, and since there will be no negative values in the probability partition arrays. However, since defining the indices of Z∞ to be the (i,j) indices in set A where both Pfm(i,j) is larger than zero and Pm(i,j) is equal to zero yields the smallest number of indices for Z∞, we will use that definition for illustration purposes since it is the least that must be done for the mathematics of FAR_Z∞ to work correctly. It will be understood that larger no-match zones may be defined, but they will include Z∞, so we illustrate the method using the smaller Z∞ definition believing that definitions of no-match zones that include Z∞ will fall within the scope of the method described.

[0196] FAR_Z∞ may be compared 22 to a desired false-acceptance-rate (“FAR”), and if FAR_Z∞ is less than or equal to the desired false-acceptance-rate, then the data set may be accepted, if false-rejection-rate is not important. If FAR_Z∞ is greater than the desired false-acceptance-rate, then the data set may be rejected 25.

[0197] If false-rejection-rate is important to determining whether a data set is acceptable, then indices in a match zone may be selected, ordered, and some of the indices may be selected for further calculations 28. Toward that end, the following steps may be carried out. The method may further include identifying a third index set ZM∞, which may be thought of as a modified Z∞, that is to say a modified no-match zone. Here we modify Z∞ so that ZM∞ includes indices that would not affect the calculation for FAR_Z∞, but which might affect calculations related to the false-rejection-rate. The indices of ZM∞ may be the (i,j) indices in Z∞ plus those indices where both Pfm(i,j) and Pm(i,j) are equal to zero. The indices where Pfm(i,j)=Pm(i,j)=0 are added to the no-match zone because in the calculation of a fourth index set (“C”), these indices should be removed from consideration. The indices of C may be the (i,j) indices that are in A but not ZM∞. The indices of C may be arranged such that

[0198] $\frac{{P_{\mathrm{fm}}\left(i,j\right)}_k}{{P_m\left(i,j\right)}_k}>=\frac{{P_{\mathrm{fm}}\left(i,j\right)}_{k+1}}{{P_m\left(i,j\right)}_{k+1}}$
to provide an arranged C index. A fifth index set (“Cn”) may be identified. The indices of Cn may be the first N (i,j) indices of the arranged C index, where N is a number for which the following is true:
FAR_Z∞∪CN=1−Σ_(i,j)∈Z∞P_fm(i,j)−Σ_(i,j)∈CNP_fm(i,j)≦FAR.

[0199] The FRR may be determined 31, where FRR=Σ_(i,j)∈CNP_m(i,j), and compared 34 to a desired false-rejection-rate. If FRR is greater than the desired false-rejection-rate, then the data set may be rejected 37, even though FAR_Z∞ is less than or equal to the desired false-acceptance-rate. Otherwise, the data set may be accepted.

[0200] FIG. 16 illustrates another method according to the invention in which the FRR may be calculated first. In FIG. 16, the reference numbers from FIG. 15 are used, but some of the steps in FIG. 16 may vary somewhat from those described above. In such a method, a first index set (“A”) may be identified, the indices in A being the (i,j) indices that have values in both Pfm(i,j) and Pm(i,j). A second index set (“Z∞”), which is the no match zone, may be identified 16. The indices of Z∞ may be the (i,j) indices of A where Pm(i,j) is equal to zero. Here we use the more inclusive definition for the no-match zone because the calculation of set C comes earlier in the process. A third index set (“C”) may be identified, the indices of C being the (i,j) indices that are in A but not Z∞. The indices of C may be arranged such that

[0201] $\frac{{P_{\mathrm{fm}}\left(i,j\right)}_k}{{P_m\left(i,j\right)}_k}>=\frac{{P_{\mathrm{fm}}\left(i,j\right)}_{k+1}}{{P_m\left(i,j\right)}_{k+1}}$
to provide an arranged C index, and a fourth index set (“Cn”) may be identified. The indices of Cn may be the first N (i,j) indices of the arranged C index, where N is a number for which the following is true:
FAR_Z∞∪CN=1−Σ_(i,j)∈Z∞P_fm(i,j)−Σ_(i,j)∈CNP_fm(i,j)≦FAR.
The FRR may be determined, where FRR=Σ_(i,j)∈CNP_m(i,j), and compared to a desired false-rejection-rate. If the FRR is greater than the desired false-rejection-rate, then the data set may be rejected. If the FRR is less than or equal to the desired false-rejection-rate, then the data set may be accepted, if false-acceptance-rate is not important. If false-acceptance-rate is important, the FAR_Z∞ may be determined, where FAR_Z∞=1−Σ_(i,j)∈Z∞P_fm(i,j). Since the more inclusive definition is used for Z∞ in this method, and that more inclusive definition does not affect the value of FAR_Z∞, we need not add back the indices where both Pfm(i,j)=Pm(i,j)=0. The FAR_Z∞ may be compared to a desired false-acceptance-rate, and if FAR_Z∞ is greater than the desired false-acceptance-rate, then the data set may be rejected. Otherwise, the data set may be accepted.

[0202] It may now be recognized that a simplified form of a method according to the invention may be executed as follows:

• • [0203] step 1: provide a first probability partition array (“Pm(i,j)”), the Pm(i,j) being comprised of probability values for information pieces in the data set, each probability value in the Pm(i,j) corresponding to the probability of an authentic match;
  • [0204] step 2: provide a second probability partition array (“Pfm(i,j)”), the Pfm(i,j) being comprised of probability values for information pieces in the data set, each probability value in the Pfm(i,j) corresponding to the probability of a false match;
  • [0205] step 3: identify a first index set (“A”), the indices in set A being the (i,j) indices that have values in both Pfm(i,j) and Pm(i,j);
  • [0206] step 4: execute at least one of the following:
    • [0207] (a) identify a first no-match zone (“Z1∞”) that includes at least the indices of set A for which both Pfm(i,j) is larger than zero and Pm(i,j) is equal to zero, and use Z1∞ to determine FAR_Z∞, where FAR_Z∞=1−Σ_(i,j)∈Z∞P_fm(i,j), and compare FAR_Z∞ to a desired false-acceptance-rate, and if FAR_Z∞ is greater than the desired false-acceptance-rate, then reject the data set;
    • [0208] (b) identify a second no-match zone (“Z2∞”) that includes the indices of set A for which Pm(i,j) is equal to zero, and use Z2∞ to identify a second index set (“C”), the indices of C being the (i,j) indices that are in A but not Z2∞, and arrange the (i,j) indices of C such that

[0209] $\frac{{P_{\mathrm{fm}}\left(i,j\right)}_k}{{P_m\left(i,j\right)}_k}>=\frac{{P_{\mathrm{fm}}\left(i,j\right)}_{k+1}}{{P_m\left(i,j\right)}_{k+1}}$

• • • [0210]  to provide an arranged C index, and identify a third index set (“Cn”), the indices of Cn being the first N (i,j) indices of the arranged C index, where N is a number for which the following is true:
      FAR_Z∞∪CN=1−Σ_(i,j)∈Z∞P_fm(i,j)−Σ_(i,j)∈CNP_fm(i,j)≦FAR
    • [0211]  and determine FRR, where FRR=Σ_(i,j)∈CNP_m(i,j), and compare FRR to a desired false-rejection-rate, and if FRR is greater than the desired false-rejection-rate, then reject the data set.
      If FAR_Z∞ is less than or equal to the desired false-acceptance-rate, and FRR is less than or equal to the desired false-rejection-rate, then the data set may be accepted. Z1∞ may be expanded to include additional indices of set A by including in Z1∞ the indices of A for which Pm(i,j) is equal to zero. In this manner, a single no-match zone may be defined and used for the entire step 4.

[0212] The invention may also be embodied as a computer readable memory device for executing a method according to the invention, including those described above. See FIG. 17. For example, in a system 100 according to the invention, a memory device 103 may be a compact disc having computer readable instructions 106 for causing a computer processor 109 to execute steps of a method according to the invention. A processor 109 may be in communication with a data base 112 in which the data set may be stored, and the processor 109 may be able to cause a monitor 115 to display whether the data set is acceptable. Additional detail regarding a system according to the invention is provided in FIG. 18. For brevity, the steps of such a method will not be outlined here, but it should suffice to say that any of the methods outlined above may be translated into computer readable code which will cause a computer to determine and compare FAR_Z∞ and/or FRR as part of a system designed to determine whether a data set is acceptable for making a decision.

[0213] If all the score producing recognition systems that comprise a biometric identification system were statistically independent, then it is well known that the joint pdf is a product of its marginal probability density functions (the “marginal densities”). It is almost certain that when we use multiple methods within one modality (multiple fingerprint matchers and one finger) or multiple biometric features of a modality within one method (multiple fingers and one fingerprint matcher or multiple images of the same finger and one fingerprint matcher) that the scores will be correlated. It is possible that different modalities are not statistically independent. For example, we may not be sure that fingerprint identification scores are independent of palm print identification scores, and we may not be sure these are different modalities.

[0214] Correlation leads to the “curse of dimensionality.” If we cannot rule out correlation in an n-dimensional biometric system of “n” recognition systems, then to have an accurate representation of the joint pdf, we would need to store an n-dimensional array. Because each dimension can require a partition of 1000 bins or more, an “n” of four or more could easily swamp storage capacity.

[0215] Fortunately, an area of mathematical statistics known as copulas provides an accurate estimate of the joint pdf by weighting a product of the marginal densities with a n-dimensional normal distribution that has the covariance of the correlated marginals. This Gaussian copula model is derived in [1: Nelsen, Roger B., An Introduction to Copulas, 2^nd Edition, Springer 2006, ISBN 10-387-28659-4]. Copulas are a relatively new area of statistics; as is stated in [1], the word “copula” did not appear in the Encyclopedia of Statistical Sciences until 1997 and [1] is essentially the first text on the subject. The theory allows us to defeat the “curse” by storing the one-dimensional marginal densities and an n-by-n covariance matrix, which transforms storage requirements from products of dimensions to sums of dimensions—or in other words, from billions to thousands.

[0216] A biometric system with “n” recognitions systems that produce “n” scores s=[s₁, s₂, . . . s_n] for each identification event has both an authentic and an impostor cdf of the form F(s₁, s₂, . . . s_n) with an n-dimensional joint density f(s₁, s₂, . . . s_n). According to Sklar's Theorem (see [1] and [2: Duss, Sarat C., Nandakumar, Karthik, Jain, Anil K., A Principled Approach to Score Level Fusion in Multimodal Biometric Systems, Proceedings of AVBPA, 2005]), we can represent the density f as the product of its associated marginals, f_i(s_i), for i=1, . . . , n, and an n-dimensional Gaussian copula function with correlation matrix R is given by
C_Rn×n(x₁, x₂ . . . x_n)=Φ_Rn×n(Φ⁻¹(x₁),Φ⁻¹(x₂) . . . Φ⁻¹(x_n))  (74)(1)

[0217] In Equation 74, x_i∈[0,1] for i=1, . . . , n, Φ is the cdf of the standard normal distribution and Φ⁻¹ is its inverse, and Φ_Rn×n is an n-dimensional Gaussian cdf whose arguments are the random deviates of standard normal distributions, that is they have a mean of zero and a standard deviation of unity. Φ_Rn×n is assumed to have covariance given by R and a mean of zero. Hence the diagonal entries of R are unity, and, in general, the off diagonal entries are non-zero and measure the correlation between components.

[0218] The density of C_Rn×n is obtained by taking its n^th order mixed partial of Equation 74:

[0219] $\begin{array}{cc}\begin{array}{c}{c}_{R_{\mathrm{nxn}}}\left(x_1,x_2\dots x_n\right)=\frac{\partial \left({\Phi }_{R_{\mathrm{nxn}}}\left({\Phi }^{-1}\left(x_1\right),{\Phi }^{-1}\left(x_2\right)\dots {\Phi }^{-1}\left(x_n\right)\right)\right)}{\partial x_1\partial x_2\dots \partial x_n}\\ =\frac{{\varphi }_{R_{\mathrm{nxn}}}\left({\Phi }^{-1}\left(x_1\right),{\Phi }^{-1}\left(x_2\right)\dots {\Phi }^{-1}\left(x_n\right)\right)}{\prod _{i=1}^n\varphi \left({\Phi }^{-1}\left(x_i\right)\right)}\left(75\right)\end{array}& \left(2\right)\end{array}$
where φ is the pdf of the standard normal distribution and φ_Rn×n is the joint pdf of Φ_Rn×n and is n-dimensional normal distribution with mean zero and covariance matrix R. Hence, the Gaussian copula function can be written in terms of the exponential function:

[0220] $\begin{array}{cc}\begin{array}{c}{c}_{R_{\mathrm{nxn}}}\left(x\right)=\frac{\frac{1}{{\left(\sqrt{2\pi }\right)}^nR}\exp \left(-\frac{1}{2}{z}^T{R}^{-1}z\right)}{{\left(\sqrt{2\pi }\right)}^n\prod _{i=1}^n{e}^{-\frac{1}{2}{z}_1^2}}\\ =\frac{1}{R}\exp \left(-\frac{1}{2}{z}^T{R}^{-1}z+\prod _{i=1}^n\frac{z_i^2}{2}\right)\left(76\right)\end{array}& \left(3\right)\end{array}$
where |R| is the determinant of the matrix R, x=[x₁, x₂, . . . x_n]^T is an n-dimensional column vector (the superscript T indicates transpose), and z=[z₁, z₂, . . . z_n]^T is an n-dimensional column vector with z_i=Φ⁻¹(x_i).
The score tuple s=[s₁, s₂, . . . s_n] of an n-dimensional biometric identification system is distributed according to the joint pdf f(s₁, s₂, . . . s_n). The pdf f can be either the authentic pdf, f_Au, or the impostor pdf, f_Im. The associated marginals are f_i(s_i), for i=1, . . . , n, which have cumulative distribution functions F_i(s_i).

[0221] Sklar's Theorem guarantees the existence of a copula function that constructs multidimensional joint distributions by coupling them with the one-dimensional marginals [1][2]. In general, the Gaussian copula function in Equation 74 is not that function, but it serves as a good approximation. So, if we set z_i=Φ⁻¹(F_i(s_i)) for i=1, . . . , n, then we can use Equation 76 to estimate the joint density as

[0222] $\begin{array}{cc}f\left(s_i,s_2,\dots s_n\right)\approx \frac{\prod _{i=1}^n{f}_i\left(s_i\right)\exp \left(-\frac{1}{2}{z}^T{R}^{-1}z+\prod _{i=1}^n\frac{z_i^2}{2}\right)}{R}\left(77\right)& \left(4\right)\end{array}$

[0223] Equation 77 is the copula model that we use to compute the densities for the Neyman-Pearson ratio. That is,

[0224] $\begin{array}{cc}\begin{array}{c}\lambda =\frac{f_{\mathrm{Au}}}{f_{\mathrm{Im}}}\\ =\frac{\frac{\prod _{i=1}^n{f}_{{\mathrm{Au}}_i}\left(s_i\right)\exp \left(-\frac{1}{2}{z}^T{R}_{\mathrm{Au}}^{-1}z+\prod _{i=1}^n\frac{z_i^2}{2}\right)}{R_{\mathrm{Au}}}}{\frac{\prod _{i=1}^n{f}_{{\mathrm{Im}}_i}\left(s_i\right)\exp \left(-\frac{1}{2}{z}^T{R}_{\mathrm{Im}}^{-1}z+\prod _{i=1}^n\frac{z_i^2}{2}\right)}{R_{\mathrm{Im}}}}\left(78\right)\end{array}& \left(5\right)\end{array}$
The densities f_Au and f_Im are assumed to have been computed in another section using non-parametric techniques and extreme value theory, an example of which appears below. To compute R_Au and R_Im we need two sets of scores. Let S_Au be a simple random sample of n_Au authentic score tuples. That is,

[0225] $S_{\mathrm{Au}}=\left\{s_{\mathrm{Au}}^1,s_{\mathrm{Au}}^2,\dots ,s_{\mathrm{Au}}^{\mathrm{\prime \prime }\mathrm{Au}}\right\},$
where s_Auⁱ is a column vector of scores from the “n” recognition systems. Likewise, let

[0226] $S_{\mathrm{Im}}=\left\{s_{\mathrm{Im}}^1,s_{\mathrm{Im}}^2,\dots ,s_{\mathrm{Im}}^{\mathrm{\prime \prime }\mathrm{Im}}\right\}$
be a simple random sample of n_Im impostor scores. We compute the covariance matrices for S_Au and S_Im with

[0227] $C_{\mathrm{Au}}=\sum _{i=1}^{n_{\mathrm{Au}}}\left(s_{\mathrm{Au}}^i-{\stackrel{\_}{s}}_{\mathrm{Au}}\right){\left(s_{\mathrm{Au}}^i-{\stackrel{\_}{s}}_{\mathrm{Au}}\right)}^T\mathrm{and}$ $C_{\mathrm{Im}}=\sum _{i=1}^{n_{\mathrm{Im}}}\left(s_{\mathrm{Im}}^i-{\stackrel{\_}{s}}_{\mathrm{Im}}\right){\left(s_{\mathrm{Im}}^i-{\stackrel{\_}{s}}_{\mathrm{Im}}\right)}^T,$
where s_Au and s_Im are the sample means of S_Au and S_Im respectively. As mentioned above, the diagonal entries of the associated R matrices are unity because the copula components are standard normal distributions, and the off diagonal entries are the correlation coefficients in the C matrices just computed. Recall that the correlation coefficient between the i^th and j^th element of C, i≠j, is given by

[0228] ${\rho }_{\mathrm{ij}}=\frac{{\sigma }_{\mathrm{ij}}}{\sqrt{{\sigma }_i\sigma j}}=\frac{C\left(i,j\right)}{\sqrt{C\left(i,i\right)C\left(j,j\right)}}.$
So, an algorithm to compute R_Au is:

• • [0229] For i=1: n_Au−1
    • [0230] For j=i+1: n_Au

[0231] $R_{\mathrm{Au}}\left(i,j\right)=\frac{C_{\mathrm{Au}}\left(i,j\right)}{\sqrt{C_{\mathrm{Au}}\left(i,i\right)C_{\mathrm{Au}}\left(j,j\right)}}$ $R_{\mathrm{Au}}\left(j,i\right)-R_{\mathrm{Au}}\left(i,j\right)$

• • [0232] For i=1: n_Au
    • [0233] R_Au(i,i)=1
      We compute R_Im in a similar manner.

[0234] The only remaining values to compute in Equation 78 are z_i=Φ⁻¹(F_i(s_i)) for each i. There are two values to compute: x_i=F_i(s_i) and then z_i=Φ⁻¹(x_i). We compute x_i=F_i(s_i) by numerically computing the integral:

[0235] $x_i={\int }_{s_{{\min }_i}}^{s_i}{f}_i\left(x\right)dx.$
Note that x_i=F_i(s_i) is the probability of obtaining a score less than s_i, hence x_i∈[0,1]. The value z_i=Φ⁻¹(x_i) is the random deviate of the standard normal distribution for which there is probability x_i of drawing a value less than z_i, hence x_i∈(−∞, ∞). We numerically solve z_i=Φ⁻¹(x_i)=√{square root over (2)}erf⁻¹(2x_i−1), where erf⁻¹(x) is the inverse error function.

[0236] Statistical modeling of biometric systems at score level may be extremely important. It can be the foundation for biometric systems performance assessment, including determination of confidence intervals, test sample size of a simulation and prediction of real world system performances. Statistical modeling used in multimodal biometric systems integrates information from multiple biometric sources to compensate for the limitations in performance of each individual biometric system. We propose a new approach for estimating marginal biometric matching score distributions by using a combination of non-parametric methods and extreme value theory. Often the tail parts of the score distributions are problematic and difficult to estimate due to lack of testing samples. Below, we describe procedures for fitting a score distribution curve. A general non-parametric method is used for fitting the center body part of a distribution curve. A parametric model, the general Pareto distribution, is used for fitting the tail parts of the curve, which is theoretically supported by extreme value theory. A biometric fusion scheme using Gaussian copula may be adapted to incorporate a set of marginal distributions estimated by the proposed procedures.

[0237] Biometric systems authenticate or identify subjects based on their physiological or behavioral characteristics such as fingerprints, face, voice and signature. In an authentication system, one is interested in confirming whether the presented biometric is in some sense the same as or close to the enrolled biometrics of the same person. Generally biometric systems compute and store a compact representation (template) of a biometric sample rather than the sample of a biometric (a biometric signal). Taking the biometric samples or their representations (templates) as patterns, an authentication of a person with a biometric sample is then an exercise in pattern recognition.

[0238] Let the stored biometric template be pattern P and the newly acquired pattern be P¹. In terms of hypothesis testing,
H₀: P=P, the claimed identity is correct;
H₁: P≠P, the claimed identity is not correct,  (38)
Often some similarity measure s=Sim(P,P¹) determines how similar patterns P and P¹ are. Decisions are made based on a decision threshold t. In most applications this threshold is fixed, i.e., t=t₀ and H₀ is decided if s≧t₀ and H₁ is decided if s<t₀.

[0239] The similarity measure s is often referred to as a score. If s is the score of a matched pair P=P′, we refer to s as a match (genuine) score; if s is the score of a mismatched pair P≠P′, we refer to s as a non-match or mismatch (impostor) score. Deciding H₀ when H₁ is true gives a false accept (FA), erroneously accepting an individual (false positive). Deciding H₁ when H₀ is true, on the other hand, results in a false reject (FR), incorrectly rejecting an individual (false negative). The False Accept Rate (FAR) and False Reject Rate (FRR) together characterize the accuracy of a recognition system, which can be expressed in terms of cumulative probability distributions of scores. The FAR and FRR are two interrelated variables and depend very much on decision threshold t₀ (see FIG. 19).

[0240] Biometric applications can be divided into two types: verification tasks and identification tasks. For verification tasks a single matching score is produced, and an application accepts or rejects a matching attempt by thresholding the matching score. Based on the threshold value, performance characteristics of FAR and FRR can be estimated. For identification tasks a set of N matching scores {s₁, s₂, . . . , s_N}, s₁>s₂> . . . >s_N is produced for N enrolled persons. The person can be identified as an enrollee corresponding to the best ranked score s₁, or the identification attempt can be rejected. Based on the decision algorithm FAR and FRR can be estimated. The usual decision algorithm for verification and identification involves setting some threshold t₀ and accepting a matching score s if this score is greater than threshold: s>t₀.

[0241] The objective of this work is to model the behavior of a biometric system statistically in terms of Probability Density Functions (PDFs). The data we present for purposes of illustrating the invention are basically two types, the biometric matching scores of genuine and impostor. In a biometric verification system, we are mostly concerned with the distribution of genuine matching scores. For an identification system, we also need to know the behavior of the impostor matching scores as well as the relation between genuine score distribution and impostor score distribution.

[0242] Applications of statistic modeling also include the following [1]:

[0243] 1. Evaluation of a biometric technique:

• • [0244] a) Distinctiveness evaluation, statistical modeling of the biometric modalities to study how well they distinguish persons absolutely.
  • [0245] b) Understanding of error probabilities and correlations between different biometric modalities.

[0246] 2. Performance prediction:

• • [0247] a) Statistical prediction of operational performance from laboratory testing.
  • [0248] b) Prediction of large database identification performance from small-scale performance.

[0249] 3. Data quality evaluation:

• • [0250] a) Quality evaluation of biometric data, relating biometric sampling quality to performance improvement.
  • [0251] b) Standardization of data collection to ensure quality; determine meaningful metrics for quality.

[0252] 4. System performance:

• • [0253] a) Modeling the matching process as a parameterized system accounting for all the variables (joint probabilities).
  • [0254] b) Determine the best payoff for system performance improvement.

[0255] The behavior of distributions can be modeled by straightforward parametric methods, for example, a Gaussian distribution can be assumed and a priori training data can be used to estimate parameters, such as the mean and the variance of the model. Such a model is then taken as a normal base condition for signaling error or being used in classification, such as a Bayesian based classifiers. More sophisticated approaches use Gaussian mixture models [2] or kernel density estimates [3]. The main limitation of all of these methods is that they make an unwarranted assumption about the nature of the distribution. For instance, the statistical modeling based on these methods weigh on central statistics such as the mean and the variance, and the analysis is largely insensitive to the tails of the distribution. Therefore, using these models to verify biometrics can be problematic in the tail part of the distribution. An incorrect estimation of a distribution especially at tails or making an unwarranted assumption about the nature of the PDFs can lead to a false positive or a false negative matching.

[0256] There is a large body of statistical theory based on extreme value theory (EVT) that is explicitly concerned with modeling the tails of distributions. We believe that these statistical theories can be applied to produce more accurate biometric matching scores. Most of the existing works that we consider as related to our work using extreme value theory are for novelty detections in medical image processing [2], [4], [5], financial risk management [6], [7], structure damage control [8], etc. Extreme values can be considered as outliers or as a particular cluster. The methods used for the detection of extreme values are based on the analysis of residuals.

[0257] This document describes a general statistical procedure for modeling probability density functions (PDFs) which can be used for a biometric system. The design of the algorithm will follow a combination of non-parametric approach and application of extreme value theory for accurate estimation of the PDFs.

[0258] A. Distribution Estimations for Biometric Systems. Statistical modeling has long been used in pattern recognition. It provides robust and efficient pattern recognition techniques for applications such as data mining, web searching, multimedia data retrieval, face recognition, and handwriting recognition. Statistical decision making and estimation are regarded as fundamental to the study of pattern recognition [7]. Generally, pattern recognition systems make decisions based on measurements. Most of the evaluation and theoretical error analysis are based on assumptions about the probability density functions (PDFs) of the measurements. There are several works which have been done for estimating error characteristics specifically for biometric systems [9]-[11].

[0259] Assume a set of matching (genuine) scores X={X₁, . . . , X_M} and a set of mismatching (impostor) scores Y={Y₁, . . . , Y_N} to be available. Here M<N because collecting biometrics always results in more mismatching than matching pairs. (From this it immediately follows that the FAR can be estimated more accurately than the FRR [10]). For the set of matching scores, X, assume that it is a sample of M numbers drawn from a population with distribution F(x)=Prob(X≦x). Similarly, let the mismatch scores Y be a sample of N numbers drawn from a population with distribution G(y)=Prob(Y≦y). The functions F and G are the probability distributions or cumulative distribution functions of match scores X and Y. They represent the characteristic behaviors of the biometric system. Since the random variables X and Y are matching scores, the distribution functions at any particular value x=t₀ can be estimated by {circumflex over (F)}(t₀) and {circumflex over (F)}(t₀) using data X and Y. In [10], [11] a simple naive estimator that is also called empirical distribution function is used:

[0260] $\begin{array}{cc}\hat{F}\left(t_0\right)=\frac{1}{M}\sum _{i=1}^M{1}_{\left\{x_i\le t_0\right\}}=\frac{1}{M}\left(\pounds X_i\le t_0\right)& \left(39\right)\end{array}$
where the estimation is simply done by counting the X_i∈X that are smaller than to and dividing by M.

[0261] In general, there are many other choices for estimating data such as matching scores. For example, among the parametric models and non-parametric models, the kernel-based methods are generally considered to be a superior non-parametric model than the naive formula in (2).

[0262] B. Another Similar Approach. In [13], another similar approach is proposed, which uses an estimate (2) for constructing the marginal distribution functions F from the observed vector of similarity scores, {X_i: i=1, 2, . . . , N}. The same empirical distribution function for the marginal distribution is written as

[0263] $\begin{array}{cc}E\left(t\right)=\frac{1}{N}\sum _{i=1}^{N}I\left(X_i\le t\right)& \left(40\right)\end{array}$
where I(A) is the indicator function of the set A; I(A)=1 if A is true, and 0 otherwise. Note that E(t)=0 for all t<t_min and E(t)=1 for all t≧t_max, where t_min and t_max, respectively, are the minimum and maximum of the observations {X_i: i=1, . . . , N}. Next, they define
H(t)≡−log(1−E(t)),
and note that discontinuity points of E(t) will also be points of discontinuity of H(t). In order to obtain a continuous estimate of H(t), the following procedure is adopted: For an M-partition [ ], the value of H(t) at a point t∈[t_min, t_max] is redefined via the linear interpolation formula

[0264] $\begin{array}{cc}H\left(t\right)=H\left(t_m\right)+\left(h\left(t_{m+1}\right)-H\left(t_m\right)\right)·\frac{t-t_m}{t_{m+1}-t_m}& \left(41\right)\end{array}$
whenever t_m≦t≦t_m+1 and subsequently, the estimated distribution function, {circumflex over (F)}(t), of {circumflex over (F)}(t) is obtained as
{circumflex over (F)}(t)=1−e^−Ĥ(t)  (42)

[0265] It follows that {circumflex over (F)}(t) is a continuous distribution function. Next they generate B*=1,000 samples from {circumflex over (F)}:

• • [0266] (1) Generate a uniform random variable U in [0,1];
  • [0267] (2) Define V=−log(1−U), and
  • [0268] (3) Find the value V* such that Ĥ(V*)=V.
    It follows that V* is a random variable with distribution function {circumflex over (F)}. To generate 1,000 independent realizations from {circumflex over (F)}, the steps (1-3) are repeated 1,000 times. Finally, a non-parametric density estimate of F is obtained based on the simulated samples using a Gaussian kernel.

[0269] Although they did not explain why the simulation or estimation procedure is so designed, we can see the underlying assumption is basically that the data follows the exponential distribution. This can be roughly seen from the following explanation. Firstly, we define an exponential distribution function by G(t)=1−e⁻¹. Then the above H(t)=−log(1−E(t))=G⁻¹(E(t)) is actually exponential quantile of the empirical distribution, which is a random variable hopefully following exponential distribution. The linear interpolation Ĥ(t) of H(t) is then an approximation of the random variable. Also {circumflex over (F)}(t) is
{circumflex over (F)}(t)=G({circumflex over (H)}(t)).

[0270] To generate simulation data, as usual, start from uniform random variable U in [0,1], data following distribution {circumflex over (F)} is then generated as
V*={circumflex over (F)}⁻¹(U)=Ĥ⁻¹G⁻¹(U).

[0271] C. The Problems. When attempting to fit a tail of a pdf using other techniques, certain problems arise or are created. They include:

• • [0272] Unwarranted assumption about the nature of the distribution.
  • [0273] Types of data used in estimation (most data are regular, the problematic data such as outliers and clusters which contribute most of the system errors are not addressed or not address enough).
  • [0274] Forgotten error analysis/estimation (decision boundaries are usually heuristic without enough understanding, which may end up with higher than lab estimated errors).

[0275] For biometric matching scores, we usually adopt that a higher score means a better match. The genuine scores are generally distributed in a range higher than the corresponding impostor scores are. For most biometric verification/identification systems, an equal error rate (EER) of 5% or lower is expected. Among such systems, those for fingerprint verification can normally get even 2% EER. For setting an optimal threshold level for a new biometric system as well as assessing the claimed performance for a vendor system, estimations of the tail distributions of both genuine and impostor scores are critical (See FIG. 19).

[0276] Most statistical methods weigh on central statistics such as the mean and the variance, and the analysis is largely insensitive to the tails of the distribution. Verification of biometrics is problematic mostly in the tail part of the distribution, and less problematic in the main body of the PDF. An incorrect estimation of a distribution especially at the tails can lead to false positive and false negative responses.

[0277] In fact, there is a large body of statistical theory based on extreme value theory (EVT) that is explicitly concerned with modeling the tails of distributions. We believe that these statistical theories can be applied to produce fewer false positives and fewer false negatives.

[0278] Although we are interested in the lower end of the genuine score distribution, we usually reverse the genuine scores so that the lower end of the distribution is flipped to the higher end when we work on finding the estimation of the distribution. Let us first concentrate on this reversed genuine matching scores and consider our analysis for estimation of right tail distributions.

[0279] To use Extreme Value Statistic Theory to estimate the right tail behaviors, we treat the tail part of the data as being realizations of random variables truncated at a displacement σ. We only observe realizations of the reversed genuine scores which exceed σ.

[0280] The distribution function (d.f.) of the truncated scores can be defined as in Hogg & Klugman (1984) by

[0281] $F_{X^{\delta }}\left(\chi \right)=P\left\{X\le \chi x/X>\delta \right\}=\{\begin{array}{cc}0& \mathrm{if}x\le \delta ,\\ \frac{F_x\left(x\right)-F_x\left(\delta \right)}{1-F_x\left(\delta \right)}& \mathrm{if}x\le \delta ,\end{array}$
and it is, in fact, the d.f. that we shall attempt to estimate. FIG. 20 depicts the empirical distribution of genuine scores of CUBS data generated on DB1 of FVC2004 fingerprint images. The vertical line indicates the 95% level.

[0282] With adjusted (reversed) genuine score data, which we assume to be realizations of independent, identically distributed, truncated random variables, we attempt to find an estimate F_Xδ(x), of the truncated score distribution F_Xδ(x). One way of doing this is by fitting parametric models to data and obtaining parameter estimates which optimize some fitting criterion—such as maximum likelihood. But problems arise when we have data as in FIG. 20 and we are interested in a high-excess layer, that is to say the extremes of the tail where we want to estimate, but for which we have little or no data.

[0283] FIG. 20 shows the empirical distribution function of the CUBS genuine fingerprint matching data evaluated at each of the data points. The empirical d.f. for a sample of size n is defined by

[0284] $F_n\left(x\right)=\frac{1}{n}\sum _{i=1}^n{1}_{\left\{X_i\le x\right\}}$
i.e. the number of observations less than or equal to x divided by n. The empirical d.f. forms an approximation to the true d.f. which may be quite good in the body of the distribution; however, it is not an estimate which can be successfully extrapolated beyond the data.

[0285] The CUBS data comprise 2800 genuine matching scores that is generated using DB1 of FVC 2004 fingerprint images (database #1 of a set of images available from the National Bureau Of Standards, fingerprint verification collection).

[0286] Suppose we are required to threshold a high-excess layer to cover the tail part 5% of the scores. In this interval we have only 139 scores. A realistic ERR level of 2% tail will leave us only 56 scores. If we fit some overall parametric distribution to the whole dataset it may not be a particularly good fit in this tail area where the data are sparse. We must obtain a good estimate of the excess distribution in the tail. Even with a good tail estimate we cannot be sure that the future does not hold some unexpected error cases. The extreme value methods which we explain in the next section do not predict the future with certainty, but they do offer good models for explaining the extreme events we have seen in the past. These models are not arbitrary but based on rigorous mathematical theory concerning the behavior of extrema.

[0287] In the following we summarize the results from Extreme Value Theory (EVT) which underlie our modeling. General texts on the subject of extreme values include Falk, Husler & Reiss (1994), Embrechts, Kluppelberg & Mikosch (1997) and Reiss & Thomas (1996).

[0288] A. The Generalized Extreme Value Distribution. Just as the normal distribution proves to be the important limiting distribution for sample sums or averages, as is made explicit in the central limit theorem, another family of distributions proves important in the study of the limiting behavior of sample extrema. This is the family of extreme value distributions. This family can be subsumed under a single parameterization known as the generalized extreme value distribution (GEV). We define the d.f. of the standard GEV by

[0289] $H_{\xi }\left(x\right)=\{\begin{array}{cc}{e}^{-{\left(1+\xi x\right)}^{\frac{1}{\xi }}}& \mathrm{if}\xi \ne 0,\\ e^{-e^{-x}}& \mathrm{if}\xi =0,\end{array}$
where x is such that 1+ξx>0 and ξ is known as the shape parameter. Three well known distributions are special cases: if ξ>0 we have the Fréchet distribution with shape parameter α=1/ξ; if ξ=0 we have the Weibull distribution with shape α=−1/ξ; if ξ=0 we have the Gumbel distribution.