Why we’re investing in neural networks and deep learning to deliver constantly improved match rates

Why we’re investing in neural networks and deep learning to deliver constantly improved match rates


When our customers invite their customers to capture identification documents, they expect GBG IDscan to automatically recognise and extract the information from that document. In order to reach high levels of automation, the dedicated research team at GBG IDscan spend as much time developing new skills as they do developing new code.

Head of Research Moneer Alitto at GBG IDscan explained: “Our team members are continuously building on their expertise especially when it comes to advanced data science, machine learning specialisations, neural networks and deep learning. The qualifications the team earn are applied straight away to tangible problems we are trying to solve for our customers.  For example, each batch of a printed document will have a slightly different positioning of the image, key fields and key security features.  If a security feature is in a slightly different place, is this due to the printing process or is it a sign that the document has been forged?  Working out what the tolerance should be for the position of fields and features on a document, with the minimum number of samples, means that our customers get new documents added to the library faster and more accurately, and can have more confidence when identifying which documents are genuine.

“A recent example is the work the research team has done to better recognise whether the document’s image has been tampered with.  Whilst on the face of it, this challenge seems relatively straightforward, it’s actually difficult for technology to “see” the same way a human eye does, especially when you consider that many document providers globally don’t print the photograph onto the document, but glue it on.  Applying their knowledge to many examples of photographs, has created data points which have enabled the team to train the technology to correctly identify tampered photographs, with a small false positive rate.  This development in technology progressed from initial concept, to prototype, through lab testing in just a few weeks.

“Our QA team is currently testing the technology against real-life scenarios that customers face, and we hope to roll it out over the summer so it can start making a positive impact to customers.”

The research team at GBG IDscan is also working on new technology to make the digital on boarding journey more robust and secure whilst still providing a great user experience. The new algorithms aim to detect whether documents are captured from display or print in order to validate the presence of physical identity documents.  This technology will add yet another layer of verification to help provide confidence that documents are genuine.

Our mission is to make it easy to on-board the right customers.  Our commitment to the learning and development of our teams is just one way that we live that mission every day.  To find out more, get in touch.