Blind watermarking method using partitioned DCT channels

A novel blind watermarking method for 8 bit grayscale images in 8/spl times/8 block DCT (discrete cosine transform) domain is presented. The method is based on an advanced spread-spectrum algorithm (see Cox, I.J. et al., NEC Research Institute Technical Report p.95-100, 1995). However, instead of correlation analysis in the detection phase, it extracts the difference between the histogram properties of pairs of subchannels. Histogram transformation is not reversible, which provides added security. Each channel is built from different DCT coefficients for the 8/spl times/8 block and the histogram describes their distribution. There are 63 AC and one DC coefficients. Channels are partitioned into some pairs of subchannels, which contain a single bit mark, thus the number of possible watermarks equals the power of 2 of the sub-channel pairs. We do not use all the channels - only the robust ones. Subchannels are generated by a key. Embedding does not change the image very much as we use a special embedding function. Watermarks currently resist JPEG compression up to QF (quality factor) 5.