Data Dialogue: Chau-Wai Wong (NCSU)
Paper Surface Based Authentication: Intrinsic Feature and Statistical Analysis
Abstract: When viewed under a microscope, mundane-seeming paper surfaces come to life, and a maze of intertwisted wood fibers creates a complicated random jungle of structures. The structures are unique for each small patch of paper and may be considered as a "fingerprint" for authentication purposes.
In this talk, I will first demonstrate that the intensity of a paper surface captured by consumer-grade cameras can be used as an authentication feature. However, such a feature is not robust to image acquisition conditions such as camera's angle and ambient light. I will then show that one can use a diffuse light reflection model to estimate an intrinsic feature of the paper, namely, the normal vector field, using camera captured photos. The normal vector field has significantly better discriminative power than using the intensity value as the feature. I will also examine a series of research questions that substantiates the practical deployment of paper-based authentication. For example, how does the discriminative power vary with respect to the patch size and digitization resolution? Can feature engineering on the estimated normal vectors yield higher discriminative power? Does the estimated normal vector resemble the real, physical quantity, or is it more of some hand-crafted feature that has no physical interpretation?