A high-capacity QRD-based blind color image watermarking algorithm incorporated with AI technologies

Abstract

In this study, a high-capacity QR decomposition (QRD) based blind watermarking algorithm with artificial intelligence (AI) technologies for color images was proposed. Watermarking implementation involves dividing the host image into non-overlapping blocks of size 4 × 4 pixels and then applying the QRD to each block. Within each block, a two-bit watermark can be embedded by manipulating the relationship between paired elements drawn from the first column of the orthogonal matrix in the ORD. Through orthonormal restoration and iterative regulation, a perfect watermark retrieval can be guaranteed in the absence of image attacks. On top of the high-capacity watermarking, the proposed algorithm also exploits two AI technologies, namely, particle swarm optimization (PSO) and super-resolution convolutional neural network (SRCNN). The PSO seeks the optimal parameters for enhancing the imperceptibility and robustness, while the SRCNN facilitates the visual recognition of extracted watermarks. In comparison with previous matrix decomposition-based watermarking algorithms, the proposed algorithm exhibits a superior performance in imperceptibility and robustness while operating at the rate of 1/8 bit per pixel. Moreover, the SRCNN refinement contributes an improvement of 0.191 to image quality in terms of mean structural similarity index measure (MSSIM).