Document identification to discourage illicit copying

An important application of future communications networks will be electronic publishing and digital library, provided copyright can be protected. A way to discourage illicit copying and distribution of documents is to uniquely mark each document copy by shifting certain lines or words slightly so that the original registered recipient can be identified from an illicit copy by detecting its mark. In this paper we present two techniques for reliable document identification, the centroid and the correlation detection. By analyzing the noise characteristics, we obtain the maximum likelihood detectors for both methods and their probabilities of error. We have applied these results to implement a marking and identification strategy proposed earlier, which marks a line both vertically by line shift and horizontally by word shift to make the marking robust against distortions in either direction. Preliminary experimental results are presented.