Determining printer and scanner resolution dependency of text classification for digital image forensics

Abstract

Forensic identification of the hardware used during printing and image scanning is a technology of value for security printing, inspection and even criminal investigations. The more familiar image forensics are concerned with determining the operations that have been performed on a digital image—usually for identifying the camera model used to capture the image. When an image is both printed and scanned, however, the forensics task is more complicated, since the print-scan (PS) cycle introduces less specific effects on the images. In order to identify the printer used to produce and read an image, classification must be performed. In this paper, we use a multi-class Adaboost classifier to determine which of 6 printers, representing 3 inkjet and 2 laserjet models, was used to produce a later-scanned image. Our results, investigating 6 different English characters, show that classification accuracy continues to increase with scanning resolution up to 1200 pixels/inch. The results are character-dependent, suggesting that different characters may be used for different forensic purposes—printer model, cartridge and individual printer identification as examples.