Minutiae template conformance and interoperability issues

Minutiae features extracted from finger images are widely used for automated fingerprint recognition. The conformance of minutiae templates to standardised data interchange formats and the interoperability of minutiae extraction and comparison subsystems from multiple suppliers are important to prevent proprietary lock-in. Based on the work performed within the European research project on minutiae template interoperability testing, this paper summarises conformance and interoperability issues that have arisen and proposes solutions. 1 Motivation for minutiae template interoperability testing Most fingerprint systems compare minutiae, i.e. characteristic points of the dermal ridges, rather than full fingerprint images. A minutia is characterised by its location, the direction of the tangent to the ridge skeleton at this point, and its type (ridge ending, ridge bifurcation, or other). To help ensure interoperability between minutiae extraction and comparison subsystems from different suppliers, an international standard for finger minutiae data interchange formats has been developed [ISO/IEC 19794-2], based on earlier national standards such as [ANSI INCITS 378, DIN V 66400]. However, in order to ensure global interoperability, the current machine-readable travel document (MRTD) specifications of the International Civil Aviation Organization (ICAO), for instance, require the storage of complete finger images, if fingerprint data is to be used in addition to a digitally stored face image, and allow minutiae data as additional option only. The use of minutiae templates instead of finger image templates would enable solutions to be memory and time saving and more privacy sensitive. If minutiae extraction and comparison subsystems from multiple suppliers are shown to be interoperable, i.e. to work well together, there could be a move from storing finger images as reference data to storing the more compact minutiae templates, even in open systems such as MRTD systems. This paper summarises issues observed within the European minutiae template interoperability testing (MTIT) project to have an impact on the interoperability of minutiae extraction and comparison subsystems from multiple suppliers and draws lessons from this. Section 2 of this paper outlines the method used for testing the conformance of minutiae templates to [ISO/IEC 19794-2] and deals with conformance issues that have been observed. Section 3 outlines the method used for testing the interoperability of subsystems and deals with interoperability issues that have been observed. Section 4 contains concluding remarks. 2 Conformance to ISO/IEC 19794-2 2.1 Conformance requirements In order to improve the chances of interoperability through standardisation, the individual subsystems within a heterogeneous system must conform to the established standards. In this spirit, conformance to standards is a prerequisite, though not a guarantee, for interoperability. [ISO/IEC 19794-2] defines several minutiae data format types. For each format type, it specifies both – syntactic requirements, characterising the structure of conforming minutiae templates, and – semantic requirements, characterising relations among fields in conforming minutiae templates or between fields in conforming minutiae templates and the underlying input finger image(s). A minutiae template conforms to a format type defined in [ISO/IEC 19794-2] only if it satisfies the relevant normative syntactic and semantic requirements for the claimed format type. A minutiae extraction subsystem is considered to conform to a format type defined in [ISO/IEC 19794-2] as long as all templates that it generates conform to that format type. 2.2 Conformance testing Conformance testing is the process of checking that a test object satisfies the conformance requirements. For an individual minutiae template as test object, conformance testing includes – format and internal consistency checking, for assessing whether the minutiae template’s fields have valid values and relate to each other as required (for instance, whether the actual length of a data field is equal to the length indicated in the corresponding length field), and – contents checking, for assessing whether the minutiae template’s fields relate to the underlying finger image(s) as required (for instance, whether minutiae are placed at the points where the dermal ridges end or bifurcate). For a minutiae extraction subsystem as test object, conformance testing involves conformance testing of the minutiae templates generated by this subsystem. The format and internal consistency checking has been conducted automatically for each minutiae template generated by the subsystem under test. This step has been conducted using decoder software that parses a minutiae template and checks it against the relevant conformance requirements. The contents checking, on the other hand, has been conducted manually for a subset of the minutiae templates generated by the subsystem under test. This step has been carried out using a minutiae visualisation tool that takes the underlying finger image as input in addition to a minutiae template. To automate also the contents checking of minutiae templates would require – a minutiae extraction reference implementation producing minutiae that are always acceptable as “ground-truth” (but there is a risk of biasing towards the chosen algorithm) or at least a database of finger images (preferably of high quality as there are no provisions for minutiae placement in low-quality images anyway) annotated with “ground-truth” reference minutiae and – a minutiae comparison reference implementation to automatically compare the minutiae templates generated by the subsystem under test with the “ground-truth” reference minutiae. 2.3 Conformance issues Conformance issues are deviations of implementations from what is required in the underlying standards. Context checking allows revealing conformance issues that would remain undetected by just format and internal consistency checking. For instance, not each minutiae extraction algorithm sets the ridge endings at the positions required by the claimed format type. Depending on the format type [ISO/IEC 19794-2] requires ridge endings to be placed either at valley bifurcation points or at ridge skeleton end points. Figure 1 shows dispersion patterns of minutiae found by four different minutiae extraction algorithms. All minutiae templates passed the format and internal consistency checking, but contents checking reveals (see Figure 1a) that the minutiae extraction algorithms place the ridge endings inconsistently (they should be placed at valley bifurcation points in the record format type). Even though minutiae comparison algorithms are quite robust against small variations, this conformance issue has a detrimental effect on the interoperability performance.