Description
AMONOT dataset integrates the use of synthetic images and aging systems to evaluate the robustness of MAD systems. In particular, this dataset enables the evaluation of how well these systems perform when dealing with aging-induced variations. The use of only synthetic images overcomes privacy limitations, and enables us to publicly release this dataset.
Aging is a very important issue in document verification since electronic Machine Readable Travel Documents (eMRTD) are typically designed to last for around 10 years. Therefore, during this time range, significant changes in a person’s facial appearance must be accepted as normal variations within the same person’s identity in verification and MAD systems.
Generation of Aged Live Images
In the generation of the AMONOT dataset, we include only ISO/ICAO compliant images, i.e. the images accepted for electronic documents. This is an important element since the MAD task is mainly focused on the analysis of images contained in electronic passports.
To simulate the effects of aging, the selected live image is artificially aged using a facial aging model. Among the others, we select HRFAE and FADING as our candidate face-aging models.
Finally, we compare the identity similarity between the original and the artificially aged images to ensure that a face recognition system is able to identify them as belonging to the same subject.
Experimental protocols
- For the evaluation of aging and aged identities, we both age and rejuvenate all ISO/ICAO images in the ONOT dataset in 10-year increments.
2. For the evaluation of MAD systems, we incrementally add 5, 10, 15, and 20 years to the subject, in order to replicate the real-world operational scenario in which an electronic document is valid for about 10 (or more) years.
Download
Click the button to download the data.
Data are organized in two folders, following the two experimental protocols described before.
For the bona fide images, please download the ONOT dataset.
For the morphed images, please download the MONOT dataset.
License and Reference
The MONOT dataset is released under the CC BY-NC 4.0 license: you may not use the material for commercial purposes and you must give appropriate credit (see the reference below). You are free to download and modify the dataset.
If you use this dataset please cite the following paper:
Spathis G., Di Domenico N., Borghi G., Franco A., Maltoni D. AMONOT: Synthetic Aging for Differential Morphing Attack Detection Systems. Proceedings of the Internation Conference on Pattern Recognition Workshops (ICPRW), 2024.