A robust image watermarking scheme using Arnold transform and BP neural network

The protection and authentication of multimedia contents and copyright have become a great concern in the fast-growing Internet environment. This paper presents an optimized robust watermarking scheme based on Arnold transform and back propagation (BP) neural network in compressed domain. Firstly, Arnold transform is improved by adding the number of variables and expanding transformation space. The security of watermark is enhanced by adding more secret keys. When the scrambled watermark is embedded into a carrier image, in order to minimize the damage to watermarked carrier image, normalization processing of watermark is added to the output of hidden layer with watermark under an established BP neural network. The compressed watermarked image is further decompressed to obtain new watermarked carrier image. In the extraction process, through dividing the watermarked image into subblocks and training the BP neural network in compressed domain again, the difference between the original and the new output of hidden layer is calculated. By using improved Arnold inverse transformation for embedding positions, the watermark coordinates are obtained with anti-normalization processing of the difference for extracting watermark. Finally, the presented algorithms have been extensively tested with different conventional signal processing and geometric attacks to verify robustness. Experimental results demonstrate that the proposed scheme has superior performance on imperceptibility and robustness over some existing algorithms with a similar approach.