Thoughts on Fingerprint Authentication

Permanent address for citation: urn:nbn:de:0125-20080325116
Contributions by Manfred Bromba (http://www.bromba.com/contacte.htm), regarding the very nature of fingerprint authentication.
# 1 Passwords, keys, and fingerprints
Updated: 2007-10-11
# 2 Are fingerprints secrets?
Updated: 2007-09-21
# 3 Widespread use of fingerprint - a danger?
Updated: 2008-06-09
# 4 Fingerprint identifiers and fingerprint verifiers
Updated: 2008-02-12

 
 

# 1 Passwords, keys, and fingerprints
First issue: 2007-10-11
It is widely accepted to divide the main personal authentication methods into knowledge ("what I know), possession ("what I have"), and biometrics ("what I am"). For a better understanding of security threats to fingerprint recognition, it should be helpful to compare these three methods to see the common features and the real differences.
1. To start with the similarities, we observe that each authentication method is based at least on a peace of information. This authentication information is to be shared between the person who wants to be authenticated and the authentication appliance which performs the authentication check. For password authentication, the password itself is the information to be checked. When using a key, the information is given by the pattern of the key bit. This pattern decides over success or failure during the authentication process which consist of inserting the key into the appropriate lock. More generally, if you are using a token like a smart card, the information is hidden in the card's chip. In fingerprint authentication, the information is the specific pattern of your finger's surface. Fingerprints may be viewed as keys with one important exception: this biological key is a fixed part of your body and cannot be removed without injury. By the way, there is also a relationship between the password which has to be learned to be effective and the behavioral part of certain biometric features, such as signature, keystroke dynamics, or voice, which also have to be trained: A password kept in mind is the limiting case of a biometric feature with 100% behavioral part! On the other hand, a password which has been written down on a piece of paper has a great relationship to a key. That is, it comprises information, tied to a piece of hardware. But in contrast to an ordinary key, the password can be taken up by any person who has a look at the sheet of paper, while the key offers its information in a form which cannot easily be learned by simple visual inspection of the key bit.
2. From a physical point of view, information cannot exist without energy. Practically, this is identical to the connection to matter. All real-life systems I know use matter to store authentication information. The kind of matter makes up the difference between passwords, keys, and fingerprint:
Authentication method Information storage for personal authentication
Password human brain
Key key hardware, usually a piece of metal
Fingerprint finger as a fixed part of the human body
If we scrutinize what kind of authentication is best suited to personal authentication, we have to state that fingerprint has a strong connection to the human body, memorized passwords have a weak connection to the human body and keys have no direct connection to the body at all. This is usually the argument to favor biometrics as personal authentication method. (On the other hand, why should one favor body over mind? Eventually, the password has the strongest connection to mind. And individuality is to a great deal determined by psychological features. But in contrast to psychological human characteristics, the password is not really a steady feature of your individuality.)
3. Authentication can only work, if the authentication information (password, key pattern, fingerprint pattern) is shared with the appliance which checks personal authorization. However, to prevent fraud, the authentication information should be kept secret against third parties which are not involved in the authentication process. If a password gets compromised, it may be used by any other person for fraud authorization. If the key pattern gets compromised, a second key may be built with exactly the same pattern to obtain a fraud authorization. The case of fingerprint is comparable to the key case, only the material is different. If the authentication information is compromised, copying of passwords is most simple. In the other two cases it is more cumbersome to copy the authentication information as it needs special tools and storage materials.
4. To increase fraud resistance, keys may be manufactured of special unique materials which are only available for authorized key manufacturers and cannot be copied by unauthorized third parties. The material composition has then to be examined with high reliability by the authentication appliance. Such a kind of problem can easily be solved by smartcards which may use cryptographic means to guarantee that the smart card has been issued by a trusted third party which will not allow unauthorized duplicates. If this method works perfectly, your authentication information may even be public. The combination between the unique authentication information and the non-copyable carrier of this information allows a better fraud resistance, if the carrier can be approved with high certainty. Unfortunately, this will only guarantee the authenticity of the token but not the authenticity of the person to be authenticated. That is, if you lose your key or card, the finder may use it for authentication in your stead - provided he knows, where the adequate lock or appliance is found.
For fingerprint, the situation is quite similar to keys. If it is possible to check the material of the finger, the authentication appliance (fingerprint sensor etc.) is able to check authenticity of the finger (and hence the person if the finger is connected to the original body). If it should prove that the finger material cannot be reconstructed artificially but can be detected as authentic and if fingerprints patterns cannot be changed artificially, the authentication information does not even need to be kept secret! Only in this case, copying of fingerprints even with assignment of the owner would not compromise authentication!
5. Certain fingerprint scanner manufacturers state that their scanners are able to distinguish between original and artificial fingers. Such statements are of little value unless the method is revealed or unless independent certifications are available which quantify the ability to distinguish between original and copy using accepted or documented measurement methods.
# 2 Are fingerprints secrets?
First issue: 2007-09-21
In his brilliant letter from August 15, 1998, titled "Biometrics: Truths and Fictions", Bruce Schneier states "Biometrics are unique identifiers, but they are not secrets". I do only partly agree with this statement. Here are my reasons:
1. I define a secret as a piece of information which is exclusively known to a limited group of secret carriers and which is kept hidden to all other parties. Here, a secret carrier may be a human person or an apparatus. My face as biometric characteristic can only be kept secret, if I keep myself away from the public or if I draw a curtain over. Furthermore, it's more and less easy for us to remember faces or to take photographs of them. In this sense I agree with Schneier, i.e., faces usually cannot be regarded as secrets. But this shouldn't be generalized to all biometric characteristics. For example, fingerprints require special efforts to become known. This effort may not be indefinitely, but the amount does not matter in the definition of a secret. There are secrets which can be easily revealed. If there is nobody to reveal them, they remain secrets. Other secrets resist even strongest attacks.
2. Unlike faces, fingerprints cannot easily be photographed using hidden cameras. Human beings have no optimized receptors for fingerprints, they can hardly remember the unique pattern of magnified fingerprint images. The reason for this may come from the fact that humans prefer and are trained to recognize each other by face, voice, size, sex, behavior etc. but not by fingerprints and not by iris, retina, or vein patterns. Nevertheless, attackers my find ways to reveal the fingerprint secret. They may do it in just the same way as it is done for passwords, for example by simulating a trusted authentication apparatus which wiretaps the biometric data without knowledge of the user. 
3. Fingerprints are left as latent prints anywhere on every surface we touch with our fingers. Latent prints remain secrets until they are uncovered. This again requires special efforts and special skills. But even professional dactyloscopists, who have a lot of experience in finding and preparing latent prints, mostly find fingerprint fragments of questionable quality which are not suitable for electronic authentication. And if they find a good fingerprint, they do not know to whom it belongs. They can find out the owner of a fingerprint only by comparing it with the fingerprints of suspicious but known persons.
4. Biometrics can be used for nearly perfect authentication, if it is possible to confirm the timely and bodily connection between biometric characteristic and the individual to avoid fraud authentications by copies. If this step can be performed with certain reliability, we call it fake detection. Only with a good fake detection, the authentication information which also has to be checked, may be public. For many biometric characteristics, this is essential. For biometric characteristics such as fingerprint, most known and realizable copy/fake detection methods do not work with sufficient reliability. In such a case, the secrecy of a biometric characteristic becomes important. This is the main, but not the only reason why fingerprint authentication works fine in practice, despite all the prophecies of doom. To summarize:
If biometric characteristics cannot be regarded as secrets, fake detection is essential
As long as a biometric characteristic can be regarded as secrets, fake detection is less essential
5. Every secret can be compromised, so do biometric secrets. Once compromised, fake detection becomes essential. Otherwise, if fake detection is imperfect, the concerned characteristic shouldn't be used any more in sensitive security applications. But there is no room for a black & white discussion. Rather, the transition between secret and public is sliding, in a statistical sense. The better the fake detection, the less effort against compromising the secret is acceptable for a constant authentication strength. If both is great, authentication strength increases! If not, fingerprint offers the chance of choosing between 10 fingers. When using certain mitigation strategies, this should be sufficient for lifetime!
6. I am aware that many security experts will strongly oppose to the idea, authentication systems could rely on biometric secrets. Finally, in the sense of security theory, only the mean amount of damage to a value counts. For specific systems, we can express this as a probability. Unfortunately, this probability cannot be determined in advance unless the complete environment and constraints are equal and measurements with numerous incidents have been made. For that reason, it will still take many years to get a final decision!
# 3 Widespread use of fingerprint - a danger?
First issue: 2007-10-11
We start with the assumption that fingerprints must be considered as secrets to allow a reliable authentication. This assumption is reasonable in view of the weak performance of present fake detection methods and the extraordinary creativity of attackers. Furthermore, we suppose that the internal technical realization of the authentication provides perfect data protection. Then hacking fingerprint verification systems without fake detection have to attack the fingerprint sensor and must overcome at least the following hurdles:
Acquire appropriate fingerprint image
Assemble a fingerprint copy suitable to fool the fingerprint acquisition system
Apply the fingerprint copy to the fingerprint acquisition system
The step Acquire is only easy for my own fingerprint. If an anti-cooperative person is attacked, this requires the skill of a dactyloscopist and delivers a set of images with often questionable quality and from which I don't know which one is most promising. The step Assemble requires knowledge of the fingerprint sensor (swipe or flat, capacitive or optical or thermal or pressure, etc.) and prepares the best material for the finger copy. Finally the step Apply tries to fool the sensor (and the system behind) with the copy made in the last step. The success of the last step depends on the number of allowed trials, on the correct fingerprint window, on the correct fingerprint angle, and on the correct treatment (pressure, speed, etc.). Altogether there are a lot of parameters which must fit. Finally, the attacker may fail due to the wrong fingerprint image. (Here we assume that a direct injection of an electronic fingerprint image into the target system is prevented by suitable and proven IT measures.)
Only the step Acquire is dedicated to the revelation of the fingerprint secret. Fingerprints may come from latent prints or may be acquired directly, e.g., by taking a photograph during sleep, by force, or using a manipulated sensor device. With an increasing number of fingerprint authentication applications, it may become unavoidable that a black sheep application collects the acquired raw images to allow its operator to access foreign systems.
As a result of the 9/11 incidents, several countries began to collect fingerprints of foreign travelers. In principle these fingerprints could be used by a secret service to assemble fingerprint copies to fool a fingerprint authentication used, e.g., for physical access control.
Phishing, which is related today to passwords, could be extended in the future to fingerprint authentication systems. Certainly, phishing with passwords is easier than extended phishing with biometrics. But this will not prevent biometrics phishing.
A further problem to be solved is that different applications show different provisions against theft of biometric data. Applications which are hardly hurt by the publicity of biometric characteristics may neglect data protection. This could seriously harm other applications where secrecy of a biometric characteristic is essential. This shows that data protection is of highest importance for all systems. In this case the weakest link in the chain may really influence the authentication strength in other systems.
Last but not least, it is to be expected that the knowhow to attack biometrics systems will become increasingly common knowledge. If biometrics would become the main authentication method, it makes more sense for a fraudulent attacker to focus on biometrics. It's simply too expensive to develop malicious tools for uncommon authentication methods. We know this effect from operating systems, browsers and credit cards: It's always the mainstream which is disproportionately high affected by attack trials. (Even if the successful attacks prevail, this does not necessarily mean that the concerned system is less protected!) 
How can we prevent biometrics from losing more and more performance? The solution sounds quite simple: Finally, with increasing dissemination, sophisticated fake detection efforts must be pushed. This would effectively impede the use of fingerprint copies. Up to then it must be guaranteed that all biometric systems fulfill highest data protection requirements.
# 4 Fingerprint identifiers and fingerprint verifiers
First issue: 2007-12-26
Most authentication systems use the concept of identifiers and verifiers. A well known example is login to a computer or network where you have to enter or select a user name ("identifier") and then to confirm your identity with a password ("verifier"). Another example is your credit card where the card with its bulky number is the identifier and your signature is the verifier. To get a deeper insight into the concept of fingerprint authentication it is useful to define an authentifier:
A (personal) authentifier is defined as the combination of all entities you need to authenticate
For example, a key, an authentication token, or the combination of user name and password may be an authentifier. Usually, an authentifier comprises a public part (e.g., a name) and a secret part (e.g., the cryptographic key within a smartcard). Biometrics even promises the chance to get along without any secret. Although identification and verification parts often are not dividable perfectly, we do so as if this would be possible and define identifier and verifier in the following way throughout this contribution:
  A (personal) identifier is a public entity which is assigned to a specific member of a closed user group such that this member can be distinguished from all other members of this group.
  A (personal) verifier is an entity which can be used to prove the claim that a person is a specific member of a closed user group.
As public we denote information which may be known to everybody without impairing the objective of an application. A piece of information is called secret when it is not known to a potential attacker.
While an identifier must be unique within the closed user group, a verifier need not when it is applied in combination with an identifier. The concept of dividing an authentifier into an identifier and a verifier has several advantages. In large user groups, identifiers become large for the sake of uniqueness. For that reason this part of an authentifier should be easy to memorize. Publicity is important for administration purposes and to initiate a contact between users. For verification, often much less information would be sufficient, e.g., a four-digit number. It has proven useful to be able to change the secret part of an authentifier independently from the identifier part. Even different levels of trust are simply manageable by using secrets of different strengths.
Normally, if two members of the user group plan an individual transaction, the knowledge of the identifiers would be sufficient. In an evil world where identity fraud is a real challenge, an identifier may not be a proof for the real identity of a user. Here, the additional use of a verifier is a solution. In compliance with our definition, verifiers may be shared secrets between users or between a user and a trusted third party which usually provides the authentication system. The use of verifiers is most important in applications where a damage to any value is to be prevented. In other words: mainly the verifier is responsible for security.
For a better understanding of the difference between identifier and verifier, let us consider two examples what can happen, when identifiers are (mis)used as verifiers. Social Security numbers (SSN) are launched in the United States to track individuals for taxation purposes. The SSN is unique for each citizen and its determination was to be an identifier. Unfortunately, certain services take the knowledge of this number as verification for the identity of a person. This is a frequent source for identity fraud since the knowledge of a public number does not testify anything about the relationship between the SSN and a person who uses it. Interestingly, instead of combating the misuse of the SSN as verifier, it is sometimes advised to keep this number secret - completely disregarding that it then would lose its manageability. 
Another example are credit cards. Credit cards have an identifier (the card number) and uses your (public) signature as verifier. Additional trust is created by using a difficult to fake card as token and sometimes using your face image as second verifier. With the emergency of the internet, this established security scheme has been flawed. Most credit card based internet transactions have no chance for a verification, instead, the public identifier, combined with certain other public information such as the name is used as verifier. By offering strong encryption for transmitting a public number, nothing but the feeling of security is simulated. The result is a rapidly increasing credit card fraud for online transactions. Up to now, the main reaction was the advice to keep the credit card data secret. As a consequence, you have to trust all shops where you have to reveal this "secret". As a side effect, the online theft of public credit card data has become more attractive. (If we call the financial damage caused by identity fraud leakage, there is no need for action, as long as leakage does not eat up the benefit for the credit card company!)
The use of fingerprints as verifiers and authentifiers becomes more and more common in security applications. Due to the open nature of fingerprints - they are not really public, but to widely accept them as secret, scientific investigations about the success rate of identity fraud with latent prints in the realistic case of non-cooperative victims are still missing - many security experts are still refusing fingerprint as a recommended method for verification. No such problems occur when using fingerprints as identifiers.
Fingerprints as identifiers. The advantage of fingerprints as identifiers is their high degree of distinguishability which expresses in a very low FAR (False Accept Rate) at moderate FRR (False Reject Rate). Using a single finger, it should be possible to separate more than 10 million members of a closed user group with a single-trial FRR of only 2% for an office worker population. (In security applications, a password or another biometric characteristic such as face, signature, or even another finger may then serve as verifier.) Faking an identifier theoretically makes little sense as security is focused on the verifier. The greatest advantage in using fingerprint as identifier is that no token is needed nor lengthy numbers or names are to be remembered and entered into a key pad. The procedure can be fast and the user cannot forget anything.
Fingerprints as verifiers. When using fingerprints a verifiers, being a part of an authentifier, this requires at least one of two assumptions to be reliable:
  The fingerprint pattern can be regarded as secret information, i.e., revealing this secret may compromise security.
  The fingerprint pattern can be regarded as public, but additional measures during capturing ensure that the fingerprint pattern does not come from a fingerprint copy.
Obviously, the easiest way would be to regard fingerprint patterns as public. This would be no problem if the capturing system comprises a copy and liveness detection. For credit card purchases in a shop copy and liveness detection is done by the seller by viewing the process of signature. Unfortunately, today no automatic fingerprint system is known which can be certified to have a lower bound for the probability resp. effort to get circumvented. Moreover, I assume that it would not be possible to ever prove such a lower limit. Does this prove fingerprint to be unusable for security? No, this situation is quite similar to encryption. Most methods used today have no proven lower limit. Even for applications using the "unbreakable" one-time pad [Wikipedia], security is not provable. Although the algorithm can be proved to be unbreakable, the problem to keep the key secret remains and is difficult to quantify. But nobody doubts about the usefulness of encryption schemes in daily life.
Secrecy of the fingerprint pattern is the most simple assumption as it would map the properties of fingerprint to an entity which is familiar from passwords. The challenge of this assumption is the availability of accepted statistical facts about how easy it is to spy out fingerprint patterns of non-cooperative persons. Otherwise, the flaws of using secrets are well known. As for ATMs [Wikipedia] and similar services I have to trust my counterparty not to misuse or to reveal my secret data. The latter requirement comes from the potential necessity to serve multiple applications with a limited stock of 10 fingerprints. From a user's perspective it seems to be reasonable to limit fingerprint authentication to a few trusted parties only. Only then, revocability of fingerprints becomes feasible. If one fingerprint is compromised, another one can be used. When using fingerprint exclusively for vocational applications this would mean that about every 5 years one fingerprint is allowed to become compromised.
For real-life systems where fingerprints are used as verifiers, one has to consider both assumptions, secrecy and publicity of the fingerprint pattern. Both parts influence fraud resistance. The effort to reveal a specific fingerprint of a non-cooperative victim is nonzero as well as the effort to successfully present a suitable copy to a fingerprint capturing device. For a security consideration, always the whole system including (non-perfect) fraud prevention measures have to be taken into account. For a successful system it is not essential to have the leakage being exactly zero. However, it is essential that leakage is smaller than the cost saving by introducing fingerprint.
When using fingerprints as verifier, distinguishability is not a problem. For that reason the recognition threshold can be reduced to the common FAR value of 0.01% for one trial. In this case, FRR can be low enough to allow at most 3 trials before temporarily disabling the user account.
Fingerprints as authentifiers. Many fingerprint identification systems on the market use fingerprint as identifier and verifier without other authentication means. This is mainly done for user convenience purposes. Since in the case of identification failures, the cause cannot clearly be attributed to a specific user who normally is represented by the identifier, such systems can only take limited measures against fraud (such as a limitation of the number of authentication trials). For that reason, only low protection requirements can be met when mainly relying on the secrecy of fingerprint patterns.
No rule without exception. So far we considered the identifier to be a unique but public information. Security was completely shifted to the function of the verifier. But this is only one side of the medal because this does not include certain attack methods which may also generate a damage. Denial of Service (DoS [Wikipedia]) is such an attack we know from web sites which are not available because of continued malicious requests. Another example is web banking. In the past, certain banks used the public account number as identifier and the password as verifier. To mitigate brute force attacks on the password, the number of unsuccessful attempts is limited (usually you reach this limit when entering the wrong password 3 times in a row). This could animate attackers to deactivate the web access to known bank accounts by simply entering a wrong password multiple times! It is at least cumbersome for the account owner to get the account reactivated. For that reason, advanced banking schemes do not use the account number but another unique number which is not known to the public!