DFIC: Towards a balanced facial image dataset for automatic ICAO compliance verification

Ensuring compliance with ISO/IEC and ICAO standards for facial images in machine-readable travel documents (MRTDs) is essential for reliable identity verification, but current manual inspection methods are inefficient in high-demand environments. This paper introduces the DFIC dataset, a novel comprehensive facial image dataset comprising around 58,000 annotated images and 2706 videos of more than 1000 subjects, that cover a broad range of non-compliant conditions, in addition to compliant portraits. Our dataset provides a more balanced demographic distribution than the existing public datasets, with one partition that is nearly uniformly distributed, facilitating the development of automated ICAO compliance verification methods. Using DFIC, we fine-tuned a novel method that heavily relies on spatial attention mechanisms for the automatic validation of ICAO compliance requirements, and we have compared it with the state-of-the-art aimed at ICAO compliance verification, demonstrating improved results. DFIC dataset is now made public (https://github.com/visteam-isr-uc/DFIC) for the training and validation of new models, offering an unprecedented diversity of faces, that will improve both robustness and adaptability to the intrinsically diverse combinations of faces and props that can be presented to the validation system. These results emphasize the potential of DFIC to enhance automated ICAO compliance methods but it can also be used in many other applications that aim to improve the security, privacy, and fairness of facial recognition systems.

💡 Research Summary

This paper addresses a critical bottleneck in the automated verification of facial images for Machine-Readable Travel Documents (MRTDs): the lack of a large-scale, diverse, and well-annotated dataset for training and evaluating models against ICAO/ISO compliance standards. Manual checking is inefficient, and existing public datasets are limited in scale, demographic balance, and coverage of non-compliant scenarios.

The authors’ primary contribution is the introduction and public release of the DFIC (Diverse Face Images - Coimbra) dataset. DFIC is a comprehensive collection comprising approximately 58,000 annotated static images and 2,706 short videos (around 5 seconds each), featuring over 1,000 unique subjects. Its key innovation lies in its intentional design for balance and comprehensiveness. Unlike previous datasets which skewed heavily towards specific demographics (e.g., young Caucasian males), DFIC ensures a more uniform distribution across gender, age (from toddlers to seniors), and ethnic origin (Asian, White/Caucasian, Other), aligning with NIST’s FRTE benchmark categories. One partition is specifically curated to be nearly uniformly distributed, facilitating bias analysis and fair model development.

The dataset meticulously covers the full spectrum of 26 photographic and pose-specific requirements outlined in the ISO/IEC 19794-5 standard (e.g., eyes closed, non-neutral expression, head rotation, occlusions, lighting issues, background clutter). It includes not only compliant portraits but also a vast array of intentionally induced non-compliant conditions using various accessories and controlled variations in pose, lighting, and expression. Half the images were captured with high-quality cameras and the other half with lower-quality devices like smartphones, reflecting real-world conditions. Each image is manually annotated for compliance/non-compliance for all 26 requirements, and when non-compliant, includes the reason and severity level. In total, DFIC provides over 2 million individual annotations, making it the largest and most richly annotated dataset of its kind.

To demonstrate the utility of DFIC, the authors employed it to fine-tune a novel automated verification method that heavily relies on spatial attention mechanisms. They evaluated this method on the established FVC-onGoing platform’s Face Image ISO Compliance Verification (FICV) benchmark, comparing it against state-of-the-art solutions like ICAONet, BioTest, and BioPass Face. The results showed improved performance (lower Equal Error Rates) across many requirements, validating that the diversity and scale of DFIC enhance model accuracy and generalization.

In summary, the DFIC dataset represents a significant leap forward for research in automated identity document validation. By providing an unprecedented volume of data with balanced demographics and exhaustive annotation of compliance violations, it directly tackles the issues of bias and limited generalization in current systems. Its public availability is poised to accelerate progress not only in ICAO compliance verification but also in broader applications aimed at improving the security, privacy, and fairness of facial recognition technologies.