Analysing Banding Features for Classifying Print Processes using Artificial Neural Networks

Abstract

The need for verifying and authenticating printed documents for the purpose of determining its validity finds applications in various fields, including digital forensic analysis and forgery detection. Print security becomes a primary concern for end use consumers and application providers at all levels of usage. Identifying the authenticity of a print can be a challenging task, since print quality and features varies on varying the methods of printing. In order to determine the source or process of printing, various methods has been specified in literature. This paper provides an in depth analysis on one such measure. Each printer has some features which are inherently part of the process due to the mechanical parameters used while modelling the device. Such features may be called intrinsic features of a printer. One of these primary features is banding that occurs primarily in electrophotographic(EP) printing processes, which has been widely reported in standard literature. However, such existence in offset or gravure printing is yet to be verified. This paper deals with examining the presence of such banding frequencies in offset or gravure printing processes. Numbers of printing processes were examined and thereby the banding was analysed using their corresponding frequency spectrum. The data obtained from the method were used as features for classifying the printing method using a supervised multilevel perceptron.