Smart Engines Patents AI Method to Detect Holographic Security Features in Documents Using Visible Light Only

Smart Engines has developed and patented a method that allows scanners to detect optically variable devices (OVD) such as holograms on identity documents using only visible light and a series of images taken under changing illumination angles.

One of the most common presentation attacks involves printing a high-quality color copy of a genuine document. While such replicas can accurately reproduce colors, text, and photographs, they cannot replicate the angle-dependent appearance of genuine OVD elements. Traditional verification relies on tilting the document by hand, but this is impossible inside a fixed scanner.

The company’s solution keeps both the document and the camera stationary. Instead, six LEDs positioned around the scanning area are switched on sequentially, producing six perfectly aligned images that differ only in lighting direction. The scanner used in experiments is the PS4-02 model manufactured by Intek Group.

Raw images first undergo dark-current subtraction. A calibration set captured from laminated white paper under the same lighting sequence is then used to normalize brightness variations caused by the illumination system itself. After normalization, pixel color vectors are projected onto a plane of average brightness, and the standard deviation across the six lighting conditions is calculated for every pixel.

Areas containing genuine holograms exhibit high standard deviation because their color changes noticeably with illumination angle. Areas without OVD remain relatively stable. The resulting deviation map is thresholded inside a predefined region of interest; if the relative area of high-deviation pixels exceeds a set limit, the document is classified as genuine.

The entire pipeline requires no new hardware or ultraviolet illumination. It demonstrates that a carefully designed algorithm can extract an additional authenticity signal from existing controlled-lighting scanners, turning six ordinary RGB frames into a reliable OVD presence detector.