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

TL;DR

This work addresses automated ICAO portrait compliance verification by introducing the DFIC dataset, a large, demographically balanced collection of 58 633 images and 2 706 videos from over 1000 subjects covering a wide range of compliant and non compliant conditions. It proposes a single model that combines a segmentation based attention mechanism with region aware classification to predict all 26 ICAO requirements in one inference. Experimental results show state of the art performance on DFIC and strong generalization to the synthetic TONO ONOT dataset, along with a formal analysis of demographic bias that highlights fairness gains when training on balanced data. The dataset and method hold practical potential for accelerating enrollment and border control workflows while improving security, privacy and fairness in facial recognition systems.

Abstract

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.

Figures (3)

• Figure 1: Examples of non-compliant images for the requirements listed in Table \ref{['table:req_list']}ferrara2012face.
• Figure 2: 24 sample pictures taken on the dataset.
• Figure 3: General architecture of the proposed method. It can be decomposed in two main branches: a segmentation branch that recurs to the DeepLabV3 architecture, resorting to a MobileNetV3 Deep Convolutional Neural Network (DCNN) as feature encoder and Atrous Spatial Pyramid Pooling (ASPP) for semantic segmentation; and a classification branch, that resorts to Mask specific Spatial Attention Modules (MsSAMs), a channel attention module through Squeeze and Excitation (SE) and a Fully Connected (FC) layer for final classification. Finally, the $\sigma$ and $\uparrow$ blocks correspond to a sigmoid activation function and an up-sampling operation, respectively.