<title>Marking and detection of text documents using transform-domain techniques</title>

Abstract

We investigate the use of frequency domain techniques to watermark text documents. A text image is essentially binary and hence contains large high-frequency components. This has several implications on the obtrusiveness and detection performance of frequency domain marking of text images, as illustrated by our extensive experiments. Generally marking is more obtrusive on a text than pictorial image. It almost always creates a 'dirty' background. 'Cleaning' the background by thresholding light grays to white renders the watermark less obtrusive but also sharply reduces the detector response, making it unrobust against noise. Both text and pictorial images seem very susceptible to shifting; this contrasts with extreme robustness against shifting of spatial domain marking through line or word shifting. Finally, we explore the combination of spatial domain marking and frequency domain detection and present preliminary experimental results on the combined approach.