Understanding CSA ISO/IEC TR 24720-14 (2019): Technical Guidance for Biometric Sample Quality in Multimodal Systems

Biometric systems increasingly rely on multimodal architectures to improve accuracy, security, and user convenience. However, ensuring consistent performance across heterogeneous data sources demands a standardised approach to sample quality assessment. CSA ISO/IEC TR 24720-14 (2019) is the Canadian adoption of the international Technical Report (TR) that provides framework-level guidance for evaluating biometric sample quality specifically in multimodal contexts. This article examines the report’s scope, technical recommendations, implementation aspects, and compliance considerations for stakeholders deploying or integrating multimodal biometric solutions.

Scope and Purpose

CSA ISO/IEC TR 24720-14 (2019) is one part of the broader ISO/IEC 24720 series on biometric sample quality. While earlier parts (e.g., Part 1) define a general framework and quality metrics, this Technical Report focuses on the unique challenges posed by systems that combine two or more biometric modalities (e.g., fingerprint, face, iris, voice). The TR provides:

• A classification of quality-related issues in multimodal fusion
• Recommendations for quality score harmonisation across modalities
• Guidelines for capturing and labelling sample quality metadata
• Best practices for using quality information in decision making (fusion, fallback, and rejection strategies)

Importantly, the document is informative — it does not prescribe mandatory requirements but rather offers a consistent reference for developers, system integrators, and test laboratories. The TR is applicable to any multimodal system operating in government, border control, banking, healthcare, or enterprise identity management.

Technical Guidance and Quality Metrics

Core Quality Dimensions

The TR identifies eight critical quality dimensions that should be assessed per sample in a multimodal pipeline:

• Character – intrinsic quality of the biometric source (e.g., scars, dryness, pose)
• Fidelity – degree of distortion, noise, or compression
• Utility – how well a sample supports the targeted recognition algorithm
• Output Quality – consistency of quality across repeated captures
• Timeliness – relevance of the quality score relative to capture time
• Interoperability – cross-vendor comparability of quality metrics
• Defect Annotation – ability to label specific sample defects
• Confidence Level – uncertainty of the quality assessment itself

Multimodal Fusion Considerations

A key contribution of the TR is its guidance on combining quality scores from different modalities. It recommends a symmetric normalisation step before any fusion to avoid dominance of one modality due to differing dynamic ranges. The report suggests three fusion strategies:

• Score‑level fusion with quality weighting – each match score is multiplied by the quality of its corresponding sample
• Decision‑level fusion with quality thresholds – a sample is rejected if its quality falls below a modality‑specific threshold
• Rank‑order fusion with quality metadata – quality information is used to break ties or reorder candidate lists

Implementation Highlights

Adopting the recommendations of CSA ISO/IEC TR 24720-14 can improve system robustness and user experience. Key implementation aspects include:

Quality Metadata Vocabulary

The TR defines a standardised set of quality metadata tags (e.g., qualityCharacterFinger, qualityFidelityFace) to be embedded in biometric data structures such as ISO/IEC 19794 records. This enables interoperability between capture devices and matching engines from different vendors.

Fallback and Rejection Logic

Multimodal systems often fall back to a single modality when others produce low‑quality samples. The TR provides decision trees for when to fall back, when to retry capture, and when to explicitly reject an attempt. It stresses that rejection should always be accompanied by a reason grounded in the quality dimension (e.g., “fingerprint quality insufficient due to character defect”).

Testing and Validation

The TR does not specify test procedures, but it recommends that implementers create a quality benchmark dataset representative of their target population. At least three scenarios should be covered: ideal conditions, degraded conditions, and adversarial manipulation attempts. The quality metrics should be validated against recognition accuracy using an independent test set.

Compliance and Adoption Notes

As a Technical Report, formal compliance is not claimed. However, many national bodies (including CSA Group in Canada) encourage adoption of TRs as pre‑normative guidance. The following points are relevant for organisations:

• Regulatory alignment: Adhering to the TR can help meet due‑diligence requirements under privacy laws (e.g., PIPEDA in Canada) by improving data quality before processing.
• Vendor conformance: Procurement specifications may reference CSA ISO/IEC TR 24720-14 to require that biometric capture devices provide standardised quality metadata.
• Audit readiness: Using the TR’s quality dimension taxonomy simplifies reporting during system evaluations and audits.

Conclusion

CSA ISO/IEC TR 24720-14 (2019) fills a critical gap by providing standardised, structured guidance for biometric sample quality in multimodal systems. While not a prescriptive standard, its taxonomy of quality dimensions, fusion recommendations, and metadata vocabulary offer a solid foundation for architects and engineers. By following the frameworks outlined in this Technical Report, organisations can build more reliable, interoperable, and privacy‑sensitive multimodal biometric solutions. For those aiming at certification or regulatory compliance, the TR should be used in conjunction with applicable normative standards.

Frequently Asked Questions

© 2026. This article is prepared for informational purposes and may contain simplified interpretations of the standard. For authoritative text, refer to the official CSA ISO/IEC TR 24720-14 (2019) publication.