SITD and across-bin matching for deformed images acquired by scanners

In this paper, a Shearing Invariant Texture Descriptor (SITD) is presented and the across-bin matching techniques Quadratic Distance (QD) and the Earth Movers Distance (EMD) are used. The shearing and 180° rotation are the main deformations generated during the image acquisition process from the physical paper using scanners. It is very common that a sheet of paper is placed imperfectly on the scanner, whereas the slight rotation is usually not based on the paper geometrical center. The acquired image therefore deformed with irregular rotation which produces a shearing transform. Besides, the image can easily be scanned upside down when the query image is acquired. This problem produces an images deformed with 180° rotation. The SITD is only invariant to the shearing transform and the 180° rotation turning positions of the extracted features. Thus, using the ordinary bin-by-bin matching techniques result in incorrect matches to the corresponding features. Hence, we proposed to use the across-bin matching techniques to perfectly match the features. The experimental results showed the superiority of the SITD along with QD when compared to the benchmark methods Local Binary Pattern (LBP) and Shearing Moment Invariant (SMI) descriptors. Besides, in most cases the QD outperformed the EMD and showed a reasonable performance.