An attack resistant hybrid blind image watermarking scheme based on combination of DWT, DCT and PCA

A blind digital image watermarking method based on combination of Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT) and Principal component analysis (PCA) is presented. First, the original image is decomposed by 3-level DWT, and HL/LH sub-band coefficients are divided into 4 x 4 non-overlapping blocks. Then, DCT is applied on each block. The high-frequency coefficients of each of four adjacent blocks are placed in a data matrix. After that, the PCA is applied to them. A binary watermarked image is scrambled by Arnold cat map and the PN-sequences of scrambled watermark bits are embedded into the first component of PCs. In the extraction procedure, the watermarked image is pre-filtered by combination of spatial sharpening and LoG filters to determine distinction between the original image and the watermarked information. Subsequently, the same method as the embedding process is used to extract the first principal component. Finally, correlation between the extracted components and the PN-sequences is calculated to decide whether the embedded bit is 0 or 1. The innovation of the proposed method is participation of PCA and frequency domain transforms for blind image watermarking purpose to improve the resistance of process against attacks, and increase the PSNR and NC even in the absence of attack. Comparing to some other methods, experimental results show high imperceptibility and the proposed method is achieved significantly higher robustness against general image processing attacks.