A Machine Learning Based Approach for Image Quality Assessment of Forged Document Images

Abstract

Document Images, such as typed and handwritten documents can be manipulated in various ways using many sophisticated digital technologies and photo editing software's. As a result, one can alter the text in the typed and handwritten documents that leads to degradation of quality of an image. The detection of multiple inherently altering operations in an image is a challenging issue, hence in this work a novel approach is proposed for the ten-class problem in which the alteration of a text can be accomplished through multiple operations, which all create the specific pattern. These operations are analysed with the help of image quality measures and classified using random forests classifier. The proposed approach gives a better classification accuracy rate of 94% for forged printed document images and 98.80% of forged handwritten document images, which is more promising and competitive with state of the art techniques reported in the literature.