Mixing fingerprints for a single biometric template
A virtual digital fingerprint image is produced by blending information from two different fingers.
Biometrics1 has made it possible to identify individuals rapidly, based on biological traits. In a biometric system, each reference template stored in the database is usually associated with only a single individual. However, in some applications, such as a joint bank account, it may be desirable to simultaneously authenticate two or more individuals associated with the account. While these individuals can be independently authenticated based on their respective biometric templates, more recent research has investigated whether a single biometric template can be generated from multiple individuals. Such an approach intricately mixes the biometric attributes of multiple individuals and generates a single joint identity. This obscures the original identities, thereby imparting privacy and security.
We have explored the possibility of generating a template representing a joint identity by a mixing process. Specifically, we create the digital identity by mixing the fingerprints of two individuals to generate a single fingerprint defining the joint identity: see Figure 1.2 A fingerprint refers to the flow of ridge patterns in the tip of the finger. The ridge flow exhibits irregularities in local regions of the fingertip, termed minutiae points. The distribution of these minutiae points, along with the associated ridge structure, is believed to be distinctive for each fingerprint.
To mix two fingerprints, the fingerprint images are represented as 2D amplitude and frequency modulated (AM-FM) signals.3 Larkin and Fletcher showed that the ridges and minutiae of a fingerprint can be completely determined by the phase of the modulated signal (i.e., the fingerprint image), while the amplitude of the signal contributes to the realistic textural appearance of the fingerprint.3 Our technique first reliably estimates the phases of the two fingerprint images to be mixed. Next, the phase of each component fingerprint image is decomposed into a continuous phase and a spiral phase.4 The continuous phase defines the ridge structure, and the spiral phase characterizes the minutiae locations (see Figure 2). Mixing is accomplished by combining the continuous phase of one fingerprint with the spiral phase of the other fingerprint (see Figure 3).5
The result is a new digital joint identity that is loosely based on the ridge structure of one fingerprint and minutiae locations of the other one. However, the mixed fingerprint is dissimilar from the two original fingerprints from the perspective of an automated fingerprint matcher. This scheme can also be used to de-identify a fingerprint by mixing it with another fingerprint (called the ‘key’) thereby obscuring the former.6, 7 Such a process is useful in applications that store the fingerprint image of an individual in a centralized database. De-identifying the image before storing it would ensure that details about the original fingerprint are obscured. Furthermore, if the database is compromised by an adversary, then a new mixed fingerprint image can be generated by merely changing the key.
Experimental results on two fingerprint databases2 demonstrated that the mixed fingerprint representing a new identity is suitable for authentication, and the mixed fingerprint is dissimilar from the original fingerprints. It also showed that the same fingerprint can be used in various applications and cross-linking between agencies can be prevented by mixing the original with a different one in each application. Mixing different fingerprints with the same fingerprint results in different identities. In addition, this approach generates a database of virtual identities from a fixed fingerprint dataset. Since mixing fingerprints can be used for de-identifying fingerprints, we also performed a detailed analysis of the security of this approach.2
In future work, we will seek to assess the viability of generating a new joint identity by combining different biometric traits of individuals. For instance, mixing a fingerprint and an iris scan generates a biometric template that inherits its distinctiveness from two (or more) different individuals. Further, we will study the possibility of using the concept of joint identity in a group authentication system.8
This project was funded by National Science Foundation Career Award Grant IIS 0642554.
Asem Othman is a PhD candidate working with Arun Ross in the Lane Department of Computer Science and Electrical Engineering.
Arun Ross is an associate professor in the Department of Computer Science and Engineering. He is co-author of the books Introduction to Biometrics and Handbook of Multibiometrics.