ISO IEC 29159-1:2010 pdf – Access to Additional Content for ISO/IEC 29159-1 First edition
ISO IEC 29159-1:2010 pdf – Access to Additional Content for ISO/IEC 29159-1 First edition.
The Type for a Type I record shall be 1 stored In I byte.
8.2.4 Distributions present
Those distribuhons present in the Type I record shall be recorded in the bits of a Distributions Present field.
8.2.5 Impostor number of comparisons
The number of comparisons used in the computation of the impostor score statistics shall be recorded as a Number of Con’toansons field.
8.2.6 Impostor distribution location
It present, this Subtvoe A field shall contain the kind. origin and value for a location parameter of the impostor score distribution If present, this value shall be followed by the scale information in 8.2.7.
NOTE If it is sensible that a normalization scheme involves only translation (i.e. no scangl. then the scale parameter otS.2.7can beset I,
8.2.7 Impostor distribution scale
If present, this Subte A field shall contain the kind, origin and value for a scale parameter of the impostor score distribution
8.2.8 Genuine number of comparisons
The number of comparisons used In the computation of the genuine score statistics shall be recorded as a Number of Comoansons field.
8.2.9 Genuine distribution location
It present, this Subtvoe A field shall contain the kind, origin and value for a location parameter of the genuine score distribution. if present, this value shall be followed by the scale information in 82. 10.
8.2.10 GenuIne distribution scale
If present, this Subtvoe A field shall contain the kind, origin and value for a scale parameter of the genuine score distribution.
8.3 Use case (Informative)
Given two Type I records, one from a fingerprint comparison subsystem (A) and another from an iris comparison subsystem (B), score-level fusion could be Implemented as the sum of z-norrnalized scores as:
Between Types 2, and 3, II must be recognized that Type 3 Is computed from essentially the data of a Type 2 record, and should be functionally equivalent. The computation of the Type 3 record will usually be parameterized to Produce a resuft more compact than Type 2. Type 3, on the other hand, requires use of a numerical method to compute the spline representation. Although source code for Type 3 is not provided here it should be preferred over Type 2 because it is more compact.
A.3 Interoperability of the types
lnteroperabllity in the FIF context refers to the use by a fusion module of. say. a Type 3 record from supplier A wilt, a Type 2 record from supplier B. Users of this part of ISO1IEC 29159 should note that Type 1 offers limited interoperability with itself and with other Types. For example a Type I record embedding moan and
standard deviation information will offer degraded performance when used with a Type 1 record embedding median and median absolute deviation Information. Further Type 1 is not readily Interoperable with Types 2 and 3. In the cases where a fusion module is provided with Type I and Type 2 records, interoperability may be achieved In two ways:
— Demotion would involve the computation of the Type I record’s statistics from the Type 2 data. For example, the median is simply the 5O° percentile. and the mean may be estimated by sampling.
Promotion would construct a Type 2 or 3 record from a Type 1. by making an assumption of a distributional form (e.g. normal), and then parameterizing it by the Type 1 location and scale parameters.
These options are not recommended, and should be considered only In cases where Type 2 or 3 data is unavailable. For these reasons it is strongly recommended that users should strongly profile this part of ISOIEC 29159 before its use. This means that a lormal document should be generated that Includes the requirement that all suppliers should generate and provide. say. a Type 2 record containing both the impostor and genuine distributions. The document might be a formal profile, or a requirements document, a RFP. or some other binding guidance.
Users should also author such profiles to support migration from one product to another. Such a profile would require suppliers to provide, say, both the comparison algorithm library and a corresponding Type 3 genuine and impostor fusion information format record.
This part of ISO/IEC 29159 has been drafted to be extensible. Specifically. a revision of the standard could include other typed records. These might establish formats to support newer, alternative or more sophisticated fusion processes (perhaps, for example, based on joint densities), or different applications (e.g. contingent fusion (8]).
A.5 Quality directed fusion
The FIF standard does not explicitly include a record for comparison score statistics conditioned on biometric sample quality measurements (21. Note, however, that a fusion module. initialized with FIF records. may nevertheless use input sample quality values as part of each subsequent fusion operation. In addition quality values might be used by a provider during the initial preparation of their FIF record.