House of cards

When used in computer technology, complex algorithms are needed to recognise and match the biometric characteristics of end users. Recently, many companies have been working towards one goal – making it possible for computers to identify humans by way of one of their unique characteristics. Either passive characteristics that may be measured (eg fingerprint, iris or retina patterns, hand geometry or DNA) or active ones where the person will have to perform a designated task (eg voice recognition, the provision of a signature or typing behaviour).

If one were to pose the question as to which of these methods is best, the answer is almost impossible to pinpoint. It will depend heavily on the application in focus.

For the end user, all methods will differ in terms of cost (mainly for the sensor), degree of user convenience, accuracy and speed. Whereas DNA analysis – for example – might be very accurate in principle, it cannot distinguish between identical twins.

Not only that, the high cost and low user convenience makes it unsuitable for day-to-day use (PC access being a good example). Would you want to give blood every time you logged on to a Windows application? Obviously not.

Voice recognition systems, on the other hand, are low cost since they simply use a standard microphone as a sensor. On the downside, they may be easily overcome by a tape-recorded voice. The same is largely true of facial recognition systems.

The system that might best be suited to IT security applications is fingerprint recognition. Nowadays the sensors are far more cost-effective, foolproof and require only a low-complexity matching algorithm.

Biometric authentication examined
How does biometric authentication work, then? In essence, the principal is the same for all methods. During a process called 'enrolment', the biometric characteristic of a user is measured and a so-called 'template' (typically of a few hundred bytes in size) computed and stored together with the user's name (or ID) in some protected store – either on a PC or smartcard. Usually, it's not possible to re-create the original measurement from this template.

During authentication, the characteristic of the user is measured again. The template is computed from the actual measurement and compared against the previously-stored (ie enrolled) template. If they are similar enough it's assumed that the user with the ID stored at the enrolment is indeed present.

A common problem with most of today's biometric authentication systems comes when an 'attacker' is able to replace the template or the ID in the database, or even alter the verification process itself. The system might recognise him or her as a different user, and subsequently grant access rights. As they exist today, this is a very real threat to all software-based access systems.

Electronic signatures have become increasingly prevalent of late. Similar to a handwritten signature which represents the will of a person on paper, the electronic signature shall do the same for electronic documents. They allow end users to identify the person concerned, ensure the integrity of e-mails, transactions and much more besides. However, they do need a secret key to be created, as well as a corresponding public key in a PKI certificate to be verified. The most common algorithm for this purpose is RSA.

Since a human being is unable to perform the complex calculations associated with an electronic signature, he or she needs to have a signature-creating device – like a computer – with the key stored in a file on the hard disk or on a smartcard that does the signing for them.

So how do these signature devices identify their rightful owner, and protect their secret key from the rest of the world? Until now, nearly all pure software solutions – as well as smartcards – have protected access to the signature key by way of a PIN. However, that PIN could intentionally (or unintentionally) be passed to some other person. Since the device accepts everyone who knows the PIN, you can never really be sure that a certain signature was created by a given person. You can only be sure that it was generated by a specific key.

Security managers can expect match-on-card technology to drive biometric authentication forward such that it becomes the standard and most convenient way of verifying users for specific IT systems

The simple way: template databases
Unfortunately, every biometric measurement – even that of the same user – creates slightly different results. Due to this the measurement itself cannot be directly used as a cryptographic key or index in some database. It must first be compared with a template for the purposes of testing similarity.

However, biometrics can be used to gain access to a signature key instead of a PIN. As it's not possible to hand over biometric characteristics to another person as easily as a PIN, signature devices that protect their key with biometrics provide security managers with a much greater certainty that only the rightful owner may activate them.

Most of today's cryptographic tokens – such as smartcards and software keyfiles (PKCS#12) – are simply protected by a PIN. Their keys provide the basics for all kinds of IT security applications and protocols. To combine them with biometric authentication, the most straightforward way is to store the PIN (together with the biometric template) either in a database or on the device itself in a readable form. If the software detects that an actual measurement fits a template, it reads the PIN from the database and presents it to the cryptographic device.

Although this appears very convenient to the user, it has a number of drawbacks for serious applications. First, the security of the device itself is still a PIN. If somebody knows that PIN, he or she can access the device without running a biometric identification. He or she may need some form of alternative application for doing so, but this is no problem for an attacker.

Second, all the PINs of the devices are stored in a database which can (in principle) be attacked, whereas before they were (hopefully) only stored in the user's memory.

Last, but by no means least, asking the user for a PIN in addition to his or her biometric characteristics eliminates the benefits of not having to remember a PIN, and gains no true security benefit as the device itself still relies solely on the PIN.

The better way: match-on-card Smartcards have become increasingly significant in the IT security market as they offer security managers a number of interesting properties. For one, the data stored on such a card is tamper-protected from any 'outside analysis'. They can use keys to perform calculations like signatures without ever revealing the key itself to the outside world, not even to the rightful owner.

And, in direct contrast to software tokens, they cannot be copied.

In essence, the cards enable the user to authenticate him or herself by sending an actual measurement of their biometric characteristic to the card instead of a PIN. The card itself will verify this, and grant access if the verification is positive.

Using such a technique provides significant advantages, both to end users and application vendors alike:

• no PIN or biometric template is stored anywhere in a database where it may be vulnerable to attack;
• privacy concerns are overcome as the biometric data is not stored outside the user's own smartcard;
• the user does not have to remember a PIN – with match-on-card this benefit can be achieved without paying the price of lower security (as it would be with the database approach, where the PIN is merely 'hidden' from the user);
• there are significant advantages for roaming users since all the data needed to authenticate the user is stored only on his or her smartcard, which is carried on their person – they do not need any form of on-line connection to some authentication server;
• existing Public Key Infrastructure (PKI) systems and security applications can be used with little or no adaptation effort – only the log-on method changes, while the rest of the cryptographic concepts remain intact (proper PKCS#11 or cryptographic service provider drivers can hide this from the application).

To achieve such a smartcard, both card and biometric vendors must work together for the good of end users. For their part, smartcard vendors need to extend the operating systems of the cards with a biometric matching algorithm supplied by biometrics specialists. Indeed, many of the larger smartcard suppliers – including Siemens and Gemplus – are currently developing such technologies.

Is match-on-card really the missing link between biometrics and cryptography? Only time will tell, but one thing is certain. End users can expect this new technology – together with the rise in smartcard use in general – to drive biometric authentication forward such that it becomes the standard and most convenient way of verifying would-be users of IT systems.