Gray-scale image watermarking using GA-BPN hybrid network

Keywords: Digital image watermarking Human visual system GA-BPN Hybrid intelligent system Robustness Discrete wavelet domain Peak-signal-to-noise ratio Normalized cross–correlation a b s t r a c t In this paper, a novel watermarking scheme is proposed by embedding a binary watermark into gray-scale images using a hybrid GA-BPN intelligent network. HVS characteristics of the images in DCT domain are used to obtain a sequence of weighting factor from a GA-BPN. This weighting factor is used to embed and extract the watermark from the image in DWT domain. The GA-BPN is trained by 27 inference rules that includes three input HVS parameters namely luminance sensitivity, edge sensitivity computed using threshold and contrast sensitivity computed using variance. The robustness of the embedding scheme is examined by executing seven different image processing attacks. Visual quality of signed images before and after the attacks is examined by PSNR. The extracted watermarks from signed and attacked images show a high degree of similarity with the embedded content. Overall, the algorithm is robust against selected attacks and is well optimized. Digital image watermarking techniques are used to establish authenticity, ownership and copyright protection of the digital content. It is primarily classified as robust and fragile. Robust watermarking of the host image deals with the enhanced possibility of recovery of the embedded content even after executing image processing attacks over the signed image. The embedding of digital content within low frequency coefficients of the image in transform domain leads to robust watermarking [16]. On the other hand, the watermarking is found to be fragile if embedding is done in high frequency coefficients. Therefore, at present, the focus of research on image watermarking is limited to three main issues. These are visual/perceptible quality of signed and attacked images, robustness and watermark embedding capacity. Among these, the former two are more significant as these are often found to be mutually exclusive. Therefore, this problem is now converged to minimize the trade off between the two. Because of this specific reason, image watermarking is presently perceived as an optimization problem. In other words, these two parameters are optimized to obtain a robust embedding and extraction scheme capable of giving good quality signed images. To this end, various soft computing intelligent techniques have been used by researchersFuzzy hybrid systems [8] and Fuzzy based Bacterial Foraging [9] have been used for this purpose. The third important parameter – watermark embedding capacity is …