29.01.2024 г.

Smart Engines has enhanced its AI system capable of detecting forged documents through holograms

The AI developer at Smart Engines has improved the functionality of its software product for recognizing and verifying the authenticity of identification documents. This system allows extracting the data from passports and ID cards in a matter of seconds, as well as verifying their authenticity, particularly by detecting holographic protection. Currently, it is one of the few systems capable of performing document verification in real-time video streams.

The holograms and other Optical variable devices (OVDs) change their appearance, such as color composition, depending on the viewing angle and lighting. Verifying the presence of any OVD on a single photograph leads to false results, by the very definition of the OVD. That is why the researchers at Smart Engines have trained the AI to verify holograms in live video streams.

This has been made possible through a proprietary method developed by the Smart Engines researchers and patented in 2019 (United States patent № US10354142B2). This method involves creating histograms of color characteristics, then comparing the histograms of the current and previous frames. Ultimately, an integrated map of hologram presence estimates is generated by combining the calculated estimates for all frames in the video stream.

The system is capable of verifying documents from all FATF countries, including Switzerland, Australia, Japan, Canada, China, Austria, Belgium, France, UK, Germany, India, Italy, Japan, Netherlands, Saudi Arabia, etc. The solution is complemented by dozens of other authenticity checks for the documents. The AI is able to detect image synthesis or interference in the document images, identifies anomalies and discrepancies related to fraudulent activities, verifies fonts, and more.

Earlier at the ICDAR-2023 scientific conference, Smart Engines scientists presented the first public dataset containing examples of identification documents with holographic security features. The MIDV-Holo dataset includes 300 video clips of documents containing holographic security features and 400 video clips simulating document presentation attacks, or, in other words, presentation of forged documents instead of original documents. Using the Smart Engines dataset, researchers worldwide will be able to test their algorithms and see how well they perform against presentation attacks perpetrated by fraudsters.

The AI developer at Smart Engines has improved the functionality of its software product for recognizing and verifying the authenticity of identification documents. This system allows extracting the data from passports and ID cards in a matter of seconds, as well as verifying their authenticity, particularly by detecting holographic protection. Currently, it is one of the few systems capable of performing document verification in real-time video streams.

The holograms and other Optical variable devices (OVDs) change their appearance, such as color composition, depending on the viewing angle and lighting. Verifying the presence of any OVD on a single photograph leads to false results, by the very definition of the OVD. That is why the researchers at Smart Engines have trained the AI to verify holograms in live video streams.

This has been made possible through a proprietary method developed by the Smart Engines researchers and patented in 2019 (United States patent № US10354142B2). This method involves creating histograms of color characteristics, then comparing the histograms of the current and previous frames. Ultimately, an integrated map of hologram presence estimates is generated by combining the calculated estimates for all frames in the video stream.

The system is capable of verifying documents from all FATF countries, including Switzerland, Australia, Japan, Canada, China, Austria, Belgium, France, UK, Germany, India, Italy, Japan, Netherlands, Saudi Arabia, etc. The solution is complemented by dozens of other authenticity checks for the documents. The AI is able to detect image synthesis or interference in the document images, identifies anomalies and discrepancies related to fraudulent activities, verifies fonts, and more.

Earlier at the ICDAR-2023 scientific conference, Smart Engines scientists presented the first public dataset containing examples of identification documents with holographic security features. The MIDV-Holo dataset includes 300 video clips of documents containing holographic security features and 400 video clips simulating document presentation attacks, or, in other words, presentation of forged documents instead of original documents. Using the Smart Engines dataset, researchers worldwide will be able to test their algorithms and see how well they perform against presentation attacks perpetrated by fraudsters.