DCT Transform Digital Watermarking is similar to spatial domain watermarking except, instead of altering the image bit plane pixel LSB, the frequency coefficients are alternated. DCT (discrete cosine transform) domain watermarking is robust against attacks such as noising, compression, sharpening, and filtering. It is based on standard JPEG compression and is suitable for applying watermarking as a parameter.
DCT allows an image to be decomposed into different frequency bands, making it easy to embed watermarking information into the frequency bands of the image. For embedding, the middle frequency bands are preferred. It increases robustness against attacks that, in many cases, may distort higher frequencies. In addition, by choosing middle frequencies, the most visually important parts of the image, concentrated in the lower frequencies, are avoided.
To prevent the extraction of the hidden information directly from the transformed domain, typically the watermarks are embedded by modifying the relationship of neighboring blocks of middle-frequency coefficients of the original image, rather than using an additive operation. Usually DCT transform digital watermarking is applied to the luminance signal.
The DCT transform is based on dividing the original image into 8 X 8 blocks of pixels and applying the transform individually to each block. The middle-frequency range from the DCT coefficients is extracted by applying a middle-frequency 2-D mask.
Assuming the digital watermark is a binary image, it is permutated by a fast 2-D pseudo random number traversing method. As a result, the spatial relationship is dispersed to resemble noise. Both pixel-based and block-based permutations are applied.
The watermark extraction procedure utilizes the original cover image and the watermark. First, the DCT transform must be performed on the watermarked and original images. Then, from the watermarked image, the original cover image must be subtracted. Both the block- and the pixel-based permutations have to be reversed to get the extracted watermark. The extracted watermark may also be scaled by a constant value.