Biometrics research could gain from open science approach to data management

Open science approaches to handling data could make biometrics research more efficient, reproducible and applicable to real-world conditions, the audience of the EAB’s latest online presentation heard.

The lunch talk on the “Desperate Need for Biometric and Identity-Related Data” was presented by Dr. Wojciech Wodo of the Wroclaw University of Science and Technology. Wodo conducts his research in collaboration with industry, in particular in payments and financial services, and he previously served as head of research and development for PayEye.

He began by introducing the concept of open science, in which methodology, sources, data, access, peer review and educational resources are all openly shared. More open data is needed in the field of biometrics and digital identity, according to Wodo. In part, this is because knowledge is dispersed, and much is held “somewhere in between domains, in between areas.”

This approach is also supported by European Commission grant programs to improve research quality, efficiency and flexibility. Repurposing previously-collected data can also speed up the pace of innovation, Wodo argues, and improve the reproducibility.

Law enforcement and border control organizations have huge collections of biometric data at their disposal, and are working on sharing them, he points out, but they are not willing or able to share that wealth of data with the private sector.

Despite the prominence of platforms like GitHub and Launchpad, more open source software and datasets are needed among industry and academic researchers to speed up the pace of innovation.

The realism needed to establish ground truth in challenging research areas, such as how to effectively block biometric injection attacks, also requires the development of an open-source biometrics repository, Wodo says. Only this amount of data can enable the accurate simulation of the noise, similarity and ambiguity faced by researchers working on problems like presentation attack detection.

Opening up, gradually

Wodo acknowledges that there are open biometric datasets, software and competitions in use today, and he points to the efforts of groups like OpenBR, as well as the EAB itself. However, he says, the maintenance of datasets and software presents a challenge to the research community.

If efforts to open up the science of biometrics were more coordinated than the independent initiatives seen so far, it would be easier to make them complete or sufficient to satisfy researchers’ needs.

General-purpose data repositories are cropping up in response to the demand for more open data, but they are not yet filled with the specific datasets that most biometrics researchers seek.

Wodo proposes the design and community endorsement of data principles based on findability, accessibility, interoperability and reusability, or FAIR. The FAIR principles could help automate searches for data, in addition to satisfying the conditions included in the acronym.

Emphasis on good data management practices is growing in the research community, Wodo says, but their implementation is still generally left to the owner of the data or the repository it is stored in.

A Forensic Informatics Biometric Repository, OpenBR, and standardization in meta-data could all contribute to a more open research environment. Publicly available data repositories and platforms could help. An integrated combination of all of the above, over time, could go much farther. Rather than storing the data itself, the meta-data could be made available for search, and data owners could still ensure regulatory compliance and ethical use. They could also receive compensation by charging a fee.

Article Topics

biometrics  |  biometrics research  |  digital identity  |  EAB  |  European Association for Biometrics  |  open science  |  open source