5. Facial image specifications

5.1 Scope

[FIPS, Section 4.4.1] requires collection of a facial image from PIV applicants, and indicates that it may be used for generation of the printed image [FIPS, Section 4.1.4.1] and for augmentation of human authentication of the card holder. The face specification in this document supports those activities, and establishes a storage format for retention of facial images. This document neither requires nor precludes agencies from retaining facial images. However, if an agency elects to retain them, then they shall be stored in the format specified here. As with other biometric elements, agencies may elect to store face data on the PIV card and use it for automated verification. Although this section places no normative requirements on such agency-optional activities, it does specify an image suited for automated biometric enrollment and face recognition.

5.2 Acquisition and format

This section provides specifications for the retention of facial images. Facial images collected during PIV Registration shall be formatted such that they conform to INCITS 385-2004 [FACESTD]. In addition to establishing a format, [FACESTD] specifies how a face image should be acquired. This is done to improve image quality and, ultimately, performance. The images shall be embedded within the CBEFF structure defined in Section 6. Because [FACESTD] is generic across applications it includes sections that have either-or requirements. Table 6 is an application profile of [FACESTD] tailored for PIV. It gives concrete specifications for much of the generic content. Column 3 references the sections of [FACESTD] and columns 4 and 5 give [FACESTD] requirements. For PIV, column 6 of Table 6 gives normative practice or value specifications. The table is not conformant with the Implementation Conformance Statement [ICS] standard. Particularly it extends the function of ICS but because it has the needed rows it may be useful in construction of a traditional ICS. Nevertheless the addition of a "values supported column" as specified in Section 9.1 of [ICS] should be used by implementers for checking conformance to the specifications.
INCITS 385 is likely to be revised by the INCITS M1 committee. Such revisions are irrelevant to PIV; however implementations should respect the version number on Line 5 of Table 6.




NORMATIVE NOTES:
1. If facial imagery is stored on the PIV Card, the length of the entire record shall fit within the container size limits specified in [800-73]. These limits apply to the entire CBEFF wrapped and signed entity, not just the [FACESTD] record. Key lengths and signing algorithms are specified in [800-78]. The size of the digital signature scales with the key length; it does not scale with the size of the biometric record.
2. More than one image may be stored in the record. It may be appropriate to store several images if appearance changes over time (beard, no beard, beard) and images are gathered at re-issuance. The most recent image shall appear first and serve as the default provided to applications.
3. When facial imagery is stored on the PIV Card, only one image shall be stored.
4. PIV facial images shall conform to the Full Frontal Image Type defined in Section 8 of [FACESTD].
5. Facial image data shall be formatted in either of the compression formats enumerated in Section 6.2 of [FACESTD]. Both whole-image and single-region-of-interest (ROI) compression are permitted. This document ([800-76]) recommends that newly collected facial image should be compressed using ISO/IEC 15444 (i.e. JPEG 2000). This applies when images will be input to automated face recognition products for authentication, and when images are stored on PIV Cards. In this latter case, ROI compression should be used. The older ISO/IEC 10918 standard (i.e. JPEG) should be used only for legacy images.
6. Facial images shall be compressed using a compression ratio no higher than 15:1. However, when facial images are stored on PIV Cards JPEG 2000 should be used with ROI compression. The innermost region should be centered on the face and compressed at no more than 24:1.
7. Face recognition performance is a function of the spatial resolution of the image. [FACESTD] does not specify a minimum resolution for the Full Frontal Image Type. For PIV, faces shall be acquired such that a 20 centimeter target placed on, and normal to, a camera's optical axis at a range of 1.5 meters shall be imaged with at least 240 pixels across it. This ensures that the width of the head (i.e. dimension CC in Figure 8 of [FACESTD]) shall have sufficient resolution for the printed face element of the PIV Card. This specification and Section 8.3.4 of [FACESTD] implies that the image width shall exceed 420 pixels. This resolution specification shall be attained optically without digital interpolation. The distance from the camera to the subject should be greater than or equal to 1.5 meters (for distortion reasons discussed in [FACESTD, Annex A.8]). The size specification is a minimum: When images are to be used for automated face recognition higher resolution is likely to yield lower error rates.
8. Facial image data shall be converted to the sRGB color space if it is stored. As stated in Section 7.4.3.3 of [FACESTD] this requires application of the color profile associated with the camera in use.