Pair digital watermarking based on unsupervised learning
She Kun, Juncai Huang, Zhou Ming-tian · 2004
A novel technique for the digital watermarking of still color images based on the concept of color space transform, wavelet fusion and image denoising is present. Algorithm transform R,G, B color space to R',G',B' color space firstly, which construct three grey images, then compute the 2Lth-level discrete wavelet decomposition of the three grey images. Embedding watermark fuse a grey image(watermark) with each of the three images' wavelet coefficients then transform R',G'B' color space to R,G,B color space to obtain the watermarked color image. When extract watermark, distill two grey images X1,X2 from the two of three wavelet coefficients, then use unsupervised ANN to denoise and get grey image(watermark) W. Test results show the superior performance of the technique and potential for the watermarking of photographic imagery.