Printed text characterization for identifying print technology using expectation maximization algorithm

Abstract

Forensic analysis of printed documents is a multi objective activity with intrinsic data as inputs which demands efficient techniques. Recent trends suggest the need for good preprocessors and post analysing tools which characterize printed text for identification of print technology. Each printing technology differs in their process of placing marking material on the target. The paper focuses on frequently used word like 'the' as test sample for characterizing printed text. The novelty of the proposed algorithm is that the selected printed text is modelled as mixture of three Gaussian models namely text, noise and background. The associated patterns and features of the models are derived using Expectation Maximization(EM) algorithm and few indices are proposed based on these parameters. One of the indices called Print Index(PI) for text is used for basic print technology discrimination. © 2011 Springer-Verlag.