Fingerprint Sensor Interoperability: Analysis of Image Quality, Minutiae Count and Performance Consistency

This research is being led by Shimon Modi & Dr. Stephen Elliott & Dr. Hakil Kim

The landscape of authentication technologies has changed drastically in the last decade. Increased use of information technology in a highly networked world has ren-dered the monolithic, stand along authentication architectures obsolete. Today’s net-worked world requires distributed authentication architecture to make it scalable and take advantage of the technological advancements. But attempting to mix disparate authentication systems raises the issue of interoperability. The effect of interoperability on the authentication process is an issue which needs to be considered when deploying distributed authentication architecture.

A typical biometric system consists of an acquisition subsystem, a feature extraction subsystem, a storage subsystem, a matching subsystem, and a decision subsystem. A fingerprint recognition system could use fingerprint sensors based on a variety of different technologies like optical, capacitive, thermal or others. The physics behind these technologies introduces distortions and variations in the captured images which are not similar. The acquisition subsystem is responsible for introducing part or all of the distortion because it is the first point of contact between the system and the subject. According to a market report by IBG, fingerprint recognition systems are the most widely deployed and commercially available biometric systems. This issue makes interoperability of fingerprint sensors a very topical one. For example, a lot of financial institutions are starting to deploy Automated Teller Machines (ATM) which use fingerprint recogni-tion for authenticating customers. Such a system can be designed to take advantage of distributed acquisition architecture and use a centralized storage and matching architecture. Without proper understanding of how fingerprints captured from different sensors affect the overall recognition rates, the financial institution would be forced to deploy the same fingerprint sensor at all the ATM’s. This requires an extraordinary level of confidence and trust in the fingerprint sensor manufacturer in order to choose just a single manufacturer. This is also a hurdle to mass absorption of this technology. If the sensor manufacturer was to stop supporting the particular fingerprint sensors, the financial institution would be forced to replace all the sensors and re-enrol all its clients. This could be a massive capital and human la-bor cost and could be a deterrent to using this technology. There is need to understand the effect of different fingerprints on recognition rates not just from an algorithm ad-vancement perspective, but also from a technology usage perspective.

The focus of this ongoing research is to understand how to lower recognition error rates for fingerprint datasets acquired from different fingerprint sensors and gain further understanding into effect of sensor specific distortions on recognition error rates.

Interoperability Testing and Analysis Framework

This study will use inferential statistics to analyze the research questions posed in this study. This study will go beyond just describing what is observed within the datasets; conclusions derived using inferential statistics can be inferred onto a representative population, which is one of the main goals of this study. The statistical analysis framework can be broken down into 3 sections:
  1. Basic fingerprint feature analysis.
  2. Match Score Analysis.
  3. Investigative Multivariate analysis.

The basic fingerprint feature analysis will involve analysis of minutiae count and image quality scores for fingerprint datasets from different sensors. Grother et al. (2004), and Fierre-Aguilar et al. (2003) have identified quality assessment as an important component of improving performance in fingerprint recognition systems. An average minutiae count and image quality score will be calculated for each fingerprint dataset corresponding to its acquisition sensor. One way analysis of variance (ANOVA) will be performed to test the differences in mean minutiae count between all the datasets.

In order to statistically analyze the differences in genuine match scores and imposter match scores between the native database and interoperability database, Dunnet’s comparison method will be used (Montgomery, 1997). Previous sensor interoperability studies have compared FNMR and FMR of native and interoperable fingerprint datasets as a method of comparison, but it does not provide any statistical basis for analysis. Dunnet’s method is a modified form of a t-test. In this method, the mean genuine match score and mean imposter match score for the native database will be considered as the control. The mean genuine match score and mean imposter match score for each interoperable database will be tested against the control (i.e. native database scores). In essence, fingerprint sensor interoperability can be described as consistency of performance of the matcher for native fingerprint databases and interoperable fingerprint databases. In statistical terms, this can be examined by testing for a significant difference of the mean genuine match scores and mean imposter match scores between native and interoperable fingerprint databases.

The purpose of the investigative analysis is to determine what kind of effect predictor variables have on the response variables. This will not be a predictive analysis, but an investigative analysis to understand and estimate the effect of these predictor variables like oiliness, elasticity, moisture content, peak pressure, and difference in quality of enrolment and verification images on the match score. Multivariate regression analysis will be performed to investigate the relationship between the predictor variables and response variable. Although this method can reveal relationships between variables, they do not imply relationships are causal. This analysis will provide insight into further analysis of relationships between variables whose contribution is significant. The match score between two fingerprint samples will be the response variable for this analysis.