An intermediate significant bit (ISB) watermarking technique using neural networks

摘要: Prior research studies have shown that the peak signal to noise ratio (PSNR) is most frequent watermarked image quality metric used for determining levels of strength and weakness watermarking algorithms. Conversely, normalised cross correlation (NCC) common after attacks were applied a verify algorithm used. Many researchers these approaches evaluate their These strategies been long time, however, which unfortunately limits value PSNR NCC in reflecting This paper considers this issue determine threshold values two parameters amount We our novel technique embedding four watermarks intermediate significant bits (ISB) six files one-by-one through replacing pixels with new and, at same keeping very close original pixels. approach gains an improved robustness based on gathered. A neural network model was built uses metrics (PSNR NCC) obtained from grey-scale images use ISB as desired output are trained each image’s NCC. The predicts together when portion or different types obtained. results indicate fluctuates before deteriorate.