DWT & SVD Based Digital Image Watermarking Using an Optimization with ANN Approach

Abstract

You can use a watermarking system to encrypt and safeguard any kind of data, from text to photos to audio and video. We decided to concentrate on image-based watermarking for security reasons. Any user with access to the source data can alter information shared on social media. Several watermarking techniques have been developed in response to this demand for secrecy. In the first, the watermark is visible to all users regardless of file type, whereas in the second, only the administrator can see it. Proposed research employs scalar value decomposition (SVD), a type of wavelet transform, and discrete wavelet transforms (DWTs) for decomposition and extraction (Singular Vector Decomposition). Consequently, our proposed approach heavily employs DWT and SVD, two decomposition methods having advantages over their competitors but also their own limits. Once the image data is collected, it can be optimised using a method like a Genetic Algorithm to produce the best results. In this study, we introduce a novel hybri Discrete Wavelet Transform (DWT) and Singular Value Decomposition (SVD) digital watermarking technique (SVD). An embedding key and watermark are added to the host image for added security and viewer verification in multimedia files. The host image and the watermark image are disassembled using a 2-level DWT strategy as part of the embedding procedure. A further subdivision of the pixel depths is achieved by the SVD technique. To further improve the stealthiness of the watermarked image, we apply a novel fitness function of genetic algorithm (GA) to the pixels generated by the SVD. Images, both unaltered and with watermarks, can now be used to teach a computer vision system (ANN). So that the watermark would be most difficult to remove, the trained network was utilised to choose the optimal area in the original image on which to place it. With this method, data loss is reduced to a minimum. Experimental results showed that the proposed method improved PSNR above that of the preceding work by Sangeetha et al (2018).