A feature calibration method for watermarking of document images

This paper describes a feature calibration scheme for use in embedding and detecting watermarks in document images. Such watermarks have a variety of uses, including copyright protection, content identification, and tamper-proofing. In general, a watermark is encoded as a displacement of certain features that can be extracted from target document images. One of the technical challenges is reliable detection of the displacement when images are distorted by print-and-scan processes. We propose a calibration method that uses the difference between a two features extracted from two sets of partitions arranged symmetrically. Since this method counter-balances the cumulative effects on the features of distortions added in the print-and-scan process, the displacement can be reliably detected. The feasibility of the method was investigated by using the average width of character strokes is used as a feature.