A block-based secure and robust watermarking scheme for color images based on multi-resolution decomposition and de-correlation

This paper presents a block-based secure and robust watermarking technique for color images based on multi-resolution decomposition and de-correlation. The principal objective of the presented scheme is to simultaneously meet all the four requirements (robustness, security, imperceptibility, and capacity) of a good watermarking scheme. The contribution of this study is to basically achieve the four contradictory requirements that a good watermarking scheme must meet. To do so, different approaches are combined in a way that the four requirements are achieved. For instance, to obtain imperceptibility, the three color channels (red, green, and blue) are de-correlated using principal component analysis, and the first principal component (de-correlated red channel) is chosen for watermark embedding. Afterwards, to achieve robustness, the de-correlated channel is decomposed using a discrete wavelet transform (DWT), and the approximate band (the other three bands are kept intact to preserve the edge information) is further decomposed into distinct blocks. The random blocks are chosen based on a random generated key. The random selected blocks are further broken down into singular values and vectors. Based on the mutual dependency on singular values and vectors’ matrices, the values are modified depending on the watermarking bits, and their locations are saved and used as another key, required when the watermark is to be extracted. Consequently, two-level authentication levels ensure the security, and using both singular values and vectors increases the capacity of the presented scheme. Moreover, the involvement of both left and right singular vectors along with singular values in the watermarking embedding process strengthens the robustness of the proposed scheme. Finally, to compare the presented scheme with the state-of-the-art schemes in terms of imperceptibility (peak signal-to-noise ratio and structural similarity index), security (with numerous fake keys), robustness (normalized correlation and bit error rate), and capacity, the Gonzalez and Kodak datasets are used. The comparison shows significant improvement of the proposed scheme over existing schemes.