Classification of Documents based on Local Binary Pattern Features through Age Analysis

Abstract

A method for identifying the age of a document using local binary pattern (LBP) features is presented here. The paper documents are the most important source of information which has been using for communication, record maintenance and proof for smoothing. A paper document is a sheet used for writing or printing something on it. We have been using a large amount of paper documents in day today life. Watermarking, embedding signatures and printed patterns have been using to secure legal documents; however, due to the misuse of digital technology, several challenges related to document security is raised. To address this problem, document age identification is one of the significant steps used to identify document’s authenticity/originality. In this paper, 500 printed documents which are published during 1993–2013 are used for identifying as new or old documents based on their year of publication. We have considered whole document for study irrespective of the content like text, line, logo, noise, etc. Initially, we have segmented a document page into $512 \times 512$ blocks and retained those text blocks which are covered with full text. Later, applied LBP technique on these text blocks and extracted features. These features are fed to k-nearest neighbors (KNN) classifier and average classification accuracy achieved is 91.5%.