Shearing invariant texture descriptor from a local binary pattern and its application in paper fingerprinting

In this paper, a Shearing Invariant Texture Descriptor (SITD) is proposed, which is a theoretically and computationally simple method based on the Rotation invariant Local Binary Pattern (Rot-LBP) descriptor. In real-world applications using flatbed scanners, such as paper texture fingerprinting, it’s common for a sheet of paper to rotate during the image acquisition process. Because the rotation is usually not based on the paper’s geometrical centre pivot, the produced image is deformed with irregular rotation resulting in shearing transforms. To tackle the shearing problem, the proposed SITD selects a few patterns from the conventional Rot-LBP to achieve either horizontal or vertical invariance. This paper presents the construction of the SITD operators and their performance in recognizing self-developed and standard image datasets, including real paper texture and Outex images, as well as those with distinctive shapes. The images were distorted with only a shearing transform. The self-developed images were distorted manually, while the standard images were distorted by software. The proposed description method achieved up to 100% correctly recognition rate in all the tested datasets based on the horizontal shear invariant operator. In addition to the accurate performance in all the conducted experiments, the operator significantly outperformed the Rot-LBP and another benchmark method, the Shearing Moment Invariant (SMI). The superiority of the descriptor in recognizing different types of patterns demonstrate its ability to be used in applications where the shearing transform is present.