واترماركينگ تصاوير ديجيتال مبتني بر ماشين بردار پشتيبان در حوزه تبديل موجك

A Robust and informed watermarking method based on SVM for coloured images using wavelet transformation

In IT management, digital media can easily be copied, modified and distributed. Copyright protection, intellectual property right protection, tracking digital media in internet space and authentication of digital media are crucial factors in solving urgent and real problem. Digital watermark methods concern as a valid solution. Most methods proposed in the literature which are robust to geometric attacks are non-blind. In this research it is proposed an algorithm for an invisible, semi-blind, robust coloured images watermarking. Existing watermarking algorithms, those exploited robust features are more or less related to the pixel position. In this method the value blue band of every pixel is used to extract some features of the host image to disregard the pixel position. Using SVM (support vector machine) the imperceptibility and robustness requirements of watermark are fulfilled and optimized. Imperceptibility is fulfilled by modifying the blue channel of the candidate pixels and surroundings in wavelet domain (LL2). Embedding has two phases, in the first phase, for every KxL block of host image, SVM l is trained according to n position selected by pseudo random number generator (PRNG) with seed1, SVMJ2 trained according to training sequence (t), generated by PRNG with seed2. In the second phase SVM l is used to select the candidate pixel for embedding at LL2 of host image and SVM 2 is used to embed the signature binary sequence by defining the value of each bit according to signatureʹs bit value Extracting watermarked signature doesnʹt need the original image and it is similar to embedding phase. Experiments show that the proposed scheme provides high PSNR of watermarked image and low error rate.