Paper under the microscopic view has a rough surface formed by intertwisted wood fibers. Such roughness is unique on a specific location of the paper and is almost impossible to duplicate. Previous work has shown that commodity scanners and cameras are capable of capturing such intrinsic roughness in term of surface normal vectors for security and forensics applications. In this paper, we examine several candidate mathematical models for camera captured images of paper surfaces and compare the modeling accuracies with reference to the measurement by the confocal microscopy. Experimental results show that the model with distinct intensity bias for images captured from different viewpoints can provide the closest result to the confocal measurement. We discover that high-frequency subbands of reconstructed 3D surfaces are more powerful than the norm map in describing the uniqueness of a physical surface. We show through a practical paper surface based authentication system that incorporating these findings can improve the discrimination performance.

Download/view the paper on IEEEXplore:

https://ieeexplore.ieee.org/document/8630759