A study of influence between digital watermarking and steganography

Researches about steganography and watermark mostly focus on either field separately. However, their correlation is ignored while they apply highly similar techniques to hide information in images. Watermark is commonly added in some websites after an image is uploaded. In this work, we study the influence of steganalysis caused by watermarking. Several watermarking and steganography algorithms, both classical and recent, are implemented and tested using an L-GEM based RBFNN. Experimental results show a degradation of steganalysis when the watermarking and steganography use similar embedding techniques. Moreover, the steganalysis accuracies depend on the use of embedding techniques in training datasets. If training images do not have watermarks while testing images have, the detection rates decrease significantly. This creates a new and easy steganography method to avoid steganalysis.