Morphing Attack Detection (MAD) by secunet
Among the types of fraud are so-called “morphing” attacks: Two biometric passport photos of different people are merged into a single facial image using image-editing software. If a person successfully submits this photo to apply for a passport, the result is a genuine but fraudulently obtained identification document that can then be used by both individuals, regardless of whether both are involved in the fraud or only one person is the driving force behind it. If the “morph” is well done, neither the facial recognition software nor the border control officer can tell the difference between the person and the morphed image.
Software algorithms that detect the described “Morpth's” during automated border control enhance border security. That is why we have been intensifying our work on MAD for several years.
secunet recently submitted the latest version of its algorithm for detecting morphed facial images to the independent and internationally recognized Face Analysis Technology Evaluation (FATE)-MORPH test conducted by the National Institute of Standards and Technology (NIST). secunet’s algorithm implements a differential approach that compares a potentially morphed facial image against a second image, which is typically captured live and is therefore trustworthy. The results of the NIST FATE-MORPH test show that the detection of morphed images has reached a performance level that enables operational use in border control. Interested readers can view the full report here.
Good results were achieved, particularly in the “High-Quality Morphs” category for example, for manually generated morphs: With a false positive rate of approximately 4% (i.e., four out of every hundred facial images are incorrectly classified as morphs and must be manually verified), the secunet algorithm detected 94% of all morphed images. With a false positive rate of approximately 1%, the MAD by secunet algorithm detected 82% of all morphed images.
The current version of the MAD by secunet algorithm can be seamlessly integrated into other secunet solutions, primarily in the areas of identity management and border control. In automated border control, via secunet easygate, it thus complements other existing security measures in the secunet border control portfolio. These solutions can provide the high-quality live images required for the differential approach as trustworthy reference images.
As a next step, secunet is continuously identifying opportunities for improvement to further optimize the performance of the MAD by secunet algorithm.
