A completed modeling of shearing and half rotation invariant texture descriptor for deformed images acquired by scanners

In this paper, a completed modeling of Shearing Invariant Texture Descriptor (SITD) is proposed and half-rotation (180° rotation) invariant features are achieved for scanners texture images applications. The main deformations generated during the image acquisition process from physical paper using flatbed scanners are shearing and half-rotation. It's very common that a sheet of paper is slightly rotated on the scanner. The acquired image is therefore deformed with irregular rotation, which produces a shearing transform. Furthermore, the image can easily be scanned upside down when the query image is acquired. This problem produces an image deformed with 180° rotation. Recently, by decomposing image local patterns into sign and magnitude components, the authors proposed the SITD only based on the first component. In this paper, we proposed a generalization approach called the Completed SITD (CSITD) employs to extract additional discrimination features based on the second component and concatenate them with their complementary from the SITD. The CSITD is however invariant only to the shearing deformation. To achieve the half-rotation invariance, a new method developed to maintain the sequence of the features of CSITD. The experimental results based on real paper texture images showed that the half-rotation invariant features of SITD (RSITD) achieved 98.1%, which is superior over the tested state-of-the-art descriptors. Implementing the half-rotation invariant method with CSITD features (CRSITD) exhibited an improvement over the RSITD with 1.9%.