Multiple Image Watermarking based on SVD : Improving Capacity and Imperceptibility

Abstract

Digital watermarking consists of inserting a mark (text, image, sound or video) in a medium. The goal is to be able to identify the author or owner of a digital document by the inserted mark. Watermarking algorithms must find a compromise between capacity and imperceptibility. Capacity represents the amount of data inserted and the imperceptibility makes the mark invisible. Our research is related to images watermarking based on singular value decomposition. There are two main approaches to improving capacity and imperceptibility : dual technique and multiple technique. Most algorithms based on these two approaches have low capacity : the watermarks are generally smaller than the host image. Agarwal & al worked on inserting an image into another image of the same size, imperceptibly. Their method adds the mark in the V matrix of the image after its singular value decomposition. A λ parameter was chosen to specify the power of insertion. Our contribution is to add two images by simultaneously improving capacity and imperceptibility. This new method consists on adding the mark in the 2 matrices U and V after singular value decomposition. The insertion of the first mark is an addition as in Agarwal’s method while that of the second mark is a subtraction to make the watermark less imperceptible. The result shows that our method is robust against different attacks such as compression, noise addition, median filtering and rotation. It is also imperceptible because not only we obtain a PSNR of 27 dB, but the histogram obtained is closer to that of the original image than that of Agarwal. A test on a database of 180 images shows that the marks are still detected. However, despite these advantages, the first mark is less visible and of poor quality compared to the second after their extraction. One solution to this is to increase the value of the parameter λ for the insertion of the first mark.