Personal identification using finger knuckles

Developing secure and effective access-control systems requires personal-identification technologies that are reliable and convenient. Hand-based biometrics exploits several internal and external features that are quite distinct in a given individual. User acceptance of hand-based biometrics systems is very high, and they are becoming more convenient and user friendly with the introduction of peg-free and touchless imaging.

The anatomy of the human hand is quite complicated. The finger-back surface—the ‘dorsum’ of the hand—can be very useful in user identification, but it has not yet attracted significant attention of researchers. In particular, the image-pattern formation from finger-knuckle bending is highly unique and thus makes this surface a distinctive biometric identifier. The anatomy of our fingers allows them to bend forward and resist backward motion. This asymmetry results in a very limited number of creases and wrinkles on their palmside. Therefore, finger-knuckle patterns are a promising avenue for further developments in touchless personal identification.¹

Figure 1. Automated extraction of finger knuckles for identification.

The advantages of employing finger-knuckle imaging are numerous. First, user acceptance of outer-palm surface imaging is very high since, unlike for fingerprints, there is no stigma of potential criminal investigation associated with this approach. Second, the finger geometry can be acquired simultaneously from the same image and employed to improve the system's performance. Peg-free imaging of the finger-back surface is also convenient. Such images can be acquired online and used to extract scale, translational, and rotational-invariant knuckle features for reliable identification.

We have developed a completely automated system that recognizes individuals using finger-knuckle images (see Figure 1). The system uses a machine-vision camera and automatically segments the relevant regions to extract meaningful information (see Figure 2). A large fraction of Indian users wear rings, signifying faith, religious belief, good health, or long-term relationships. The system has therefore been designed to automatically identify the presence of rings and extract the most suitable region for finger identification.

Texture analysis of the normalized knuckle-image regions can reveal highly discriminative information for identification purposes. Analysis of the acquired knuckle images in both the spatial and frequency domains has also been explored. Two-dimensional Gabor filters are appropriate for this purpose:² they have tunable angular- and axial-frequency bandwidths and center frequencies, and achieve optimal joint resolution in both the spatial and frequency domains. Phase information can be extracted from knuckle creases and lines using comparative responses from the even and odd components of the Gabor filters and used to form a ‘KnuckleCode,’ similar to IrisCode² or FingerCode.³ Alternatively, the dominant orientations of Gabor filters in a filter bank can also be used to extract phase information.⁴ Personal identification using such a KnuckleCode yields promising results that are comparable to or better than several other current biometric approaches.^1,5 A comparative performance study using individual knuckle images from the five fingers of one hand suggests that middle and ring fingers yield highly discriminant information and achieve the best performance compared to the thumb, index, or little finger.

The performance of finger-knuckle identification depends sensitively on the accuracy of knuckle segmentation from the fingers or hands being measured. Therefore, further performance improvement can be achieved by developing more accurate knuckle-segmentation schemes. This can also be achieved through some tradeoff in user convenience by employing pegs during imaging (as done in some earlier versions of hand-geometry or palmprint systems).

Figure 2. Live display and hand recognition.

In summary, the first online personal-identification system employing finger-knuckle surface measurements has achieved promising results and an accuracy comparable to or better than other hand-based biometrics systems. However, efforts to employ traditional texture-phase information using knuckle lines and creases are not yet satisfactory and further efforts are required to investigate the performance of knuckle biometrics for potential application to a large user population.