基於DWT-SVD-DCT轉換的數位浮水印嵌入技術

Abstract

In the information age, a large amount of digital image information can be acquired from the Internet. However, any unauthorized tampering may seriously jeopardize the legitimate interests of the digital information owner, and the protection of intellectual property rights has become an important issue. Hence, image watermarking is used to resolve the intellectual property contentious issues. For many transform-domain watermarking techniques, attempts have been made to exploit various transformations to extract image features, and then embed the watermark into those feature parameters. This study used an innovative digital watermarking technology constructed by integrating the discrete wavelet transform (DWT), singular value decomposition (SVD) and discrete cosine transform (DCT). The transformation technologies of DWT, SVD, and DCT have their respective advantages. DWT can resolve the low frequency image, SVD can capture the geometric textures of various layers, and DCT can be used to analyze the components of specific frequencies. This study first identified the feature parameters of the composite DWT-SVD-DCT domain, and then used the quantization index modulation (QIM) rule to embed the watermark. To verify the effect of the proposed scheme, the resulting performance were compared with five other watermarking technologies. The performances of the watermarking technologies will evaluated by embedding a 64x64 binary watermark into five 512x512 greyscale images. The results showed that the proposed scheme can result in a PSNR value up to 42 dB after watermark embedding. For a watermarked image with a PSNR greater than 40dB, the differences between the watermarked and original image are almost undetectable by human eyes. The proposed scheme also exhibited satisfactory robustness against common digital signal processing attacks. Even under the high-intensity attack of JPEG compression (e.g., quality factor = 20), the average mean of the normalized correlation (NC) values of the recovered watermarks still reached 0.889. Moreover, the proposed method can effectively resist the general attacks by Gaussian noise, salt and pepper noise, filtering, rotation, cropping, and brightness adjustment.