Partial Automatic Verification of the Compliance of ICAO requirements for Portraits

Biometric verification plays an extremely important role in today’s context of international travel and identification of humans. In this regard, the International Civil Aviation Organization (ICAO) has created a document based on the ISO/IEC 19749-5 standard, aimed at ensuring compliance with the photographic requirements of facial images for use in official documents. However, this verification process is still largely carried out manually by qualified professionals, an approach that is not only subjective but also time-consuming. There is a growing need to develop automatic verification systems for this purpose.

This dissertation seeks to address this challenge by exploring approaches based on the use of neural networks – YOLOv8 and ResNet-50 – to verify a facial photograph according to ICAO requirements, specifically those related to the verification of prescription glasses. Both architectures were trained and tested using a dataset resulting from a partnership between the ISRCoimbra (Institute of Systems and Robotics) and the INCM (”Imprensa Nacional Casa da Moeda”) as part of the FACING2 project, which aims to create a dataset of facial photographs to assess compliance with ICAO requirements.

The proposed solutions were developed based on the classification models resulting from the use of YOLOv8 and ResNet neural networks, with the aim of automating the verification of ICAO requirements. Finally, the results obtained are discussed and compared with a method in the literature called ICAONet. The results are quite promising, as they are on par with some of the best results published by this methodology, suggesting that neural network-based approaches have the potential to significantly improve the efficiency and accuracy of biometric verification in accordance with ICAO standards.