Researchers explore morph attacks on contactless fingerprint biometrics

In today’s smartphone market, almost every new device comes equipped with some form of biometric authentication for enhanced security. While contactless fingerprint biometrics have been in use for some time, there is a growing concern about the susceptibility of smartphone-based fingerprint biometric systems to morphing attacks.

A research team from the Norwegian University of Science and Technology has put forward a method to generate morphing attacks on “fingerphoto” biometrics captured using smartphones. They have developed three distinct algorithms for generating morphing attacks at the image level, allowing them to create morphed fingerprint images with minimal distortions.

The research uses two datasets: one collected using an iPhone 6s in indoor conditions and another collected from the publicly available dataset IIITD (captured using iPhone 5s). The proposed morphing algorithms are shown to be vulnerable to commercial off-the-shelf (COTS) and block-directional fingerprint verification (BDFV) systems. The researchers chose to test Neurotechnology’s VeriFinger SDK as a commercial system with “fast and reliable fingerprint matching performance.”

The research draws on previous work on face morphing conducted by biometrics experts Kiran Raja and Christoph Busch. Busch emphasizes that biometric recognition systems with higher accuracy are often more susceptible to morphing attacks than less accurate ones, as demonstrated in this research paper.

The three algorithms function through a sequence of steps, including pre-processing, triangulation, warping, and blending. Each algorithm differs based on the keypoint detection methods used – accelerated segment test (FAST), scale-invariant feature transform (SIFT), and central point of the grid.

In the initial pre-processing step, “fingerphoto” images undergo segmentation and region of interest (ROI) extraction. The images, captured using a smartphone, are utilized to extract the ROI by implementing techniques such as background cropping, alignment, and enhancement using methods like Frangi filters.

During the triangulation step, the algorithms generate “reliable” morphs by dividing the fingerphoto images into rectangular grids of varying sizes. Keypoints within each grid are identified using the three aforementioned methods. These keypoints are then used to construct triangles within each grid, forming the basis for the morphing process.

The wrapping stage involves converting the triangles from the original images into the corresponding triangles in the morphed images. The final step involves blending the wrapped images to create a morphed image.

The study also proposes a morphing attack detection algorithm utilizing handcrafted and deep features. However, the researchers noted that the proposed model encountered challenges in accurately detecting this type of attack, as indicated by the high error rates in the detection results.

Although both COTS and BDFV systems are susceptible, the commercial software has a higher potential for attack. “The vulnerability of the morphing images also depends on the quality of the fingerphoto samples used to generate morphing,” the research paper concludes.

Article Topics

biometrics  |  biometrics research  |  fingerphoto biometric  |  fingerprint biometrics  |  morphing attack  |  smartphones  |  spoofing