LBP-based degraded document image binarization

Abstract

Most of the classical threshold-based methods for document image binarization use simple features carried out from the spatial pixels values of the document images. In this paper, we present a new binarization method for degraded documents, based on Local Binary Pattern (LBP) as a texture measure. The mean and variance of pixels are computed respectively from both the original document image and the LBP image. Then, these features are used within a threshold-based method. Another variant is computed by combining a contrast information with the LBP operator to overcome the drawback caused by the poor contrasted document images. Experimental results conducted on DIBCO datasets and compared against some state-of-the-art methods, prove the effective use of the LBP for binarizing historical documents.