Non-blind Image Watermarking Algorithm based on Non-Separable Haar Wavelet Transform against Image Processing and Geometric Attacks

Abstract

The protection of digital media become crucial and needs to be developed as technology continues to become more advanced. This is to cater to the problems in copyright protection and data security. This paper presents a new non-blind image watermarking algorithm applying modified non-separable Haar wavelet transform (NSHWT), singular value decomposition (SVD), Arnold’s cat map and Rabin-p cryptosystem. The traditional transform domain watermarking (DWT) is resource-consuming, especially when performed on large image data. In order to improve on this issue, the proposed algorithm applied the modified NSHWT, a much more efficient alternative to the DWT. Another concern of digital watermarking algorithms is it is often publicly known, so encryptions are important to keep the image secure. The embedded watermark image is protected by scrambling the image using Arnold’s cat map, and the scrambling parameters are encrypted with the Rabin-p cryptosystem to keep the algorithm secure. In general, the application of modified NSHWT and SVD makes the watermark highly robust against image processing and geometric processing attacks, and its security is ensured with the protection by Arnold’s cat map and Rabin-p cryptosystem. There are already many algorithms with high imperceptibility and robustness, but in finding the perfect balance, authors often forsake other metrics such as efficiency, security, and flexibility. These are also essential when the algorithm is considered in real applications. Thus, the proposed algorithm is evaluated in terms of imperceptibility, robustness, and efficiency. The algorithm achieved imperceptibility results of 51.5157 average PSNR and 0.9991 average SSIM, and robustness results from 0.9710 to 0.9966 NC values. The algorithm is also around 40% faster to execute with modified NSHWT compared to the traditional DWT. Finally, it has a high embedding capacity of 6 bits per pixel. Overall, results show that the algorithm is highly imperceptible and robust against image processing and geometric attacks, while additionally being secure and flexible.