A robust image watermarking scheme invariant to rotation, scaling and translation attacks

Abstract

Digital watermarking has been proposed as an effective means of copyright protection, copy protection and finger - prin ting of digital multimedia. These are all means of protecting the digital multimedia from illegal copying and/or distribution. The need for protection of digital multimedia is much more acute than their analogue counterparts. This is because they are trans mitted over the internet exposing them to the risk of being copied illegally. Various methods of digital image watermarking have been proposed. Attackers always attempt to remove the embedded watermarks or render them useless by using either geometric atta cks or signal processing attacks. Most watermarking algorithms are fairly robust to signal processing attacks but are fragile to geometric attacks. This study proposes a watermarking scheme that is robust to geometric distortion attacks and signal processi ng attacks without increasing the computational complexity of the algorithm. The proposed method is based on embedding the watermark in the Discrete Cosine Transform (DCT) domain and then employing a spread - spectrum format to enhance security and Vector Qu anti zation (VQ) techniques to compress the image. Since embedding the watermark in the DCT domain gives it sufficient robustness to signal processing attacks, the proposed techniq ue focuses more on rotation, scaling and translation ( RST ) attacks . This meth od employs the Harris corner detector - based feature - points to obtain a Delaunay tessellation that is used in reversing the geometric attacks before attempting to extract the watermark. In situations where the RST attacks lead to formation of substantial ly dark regions in the image, some feature - points are lost or additional feature - points are obt ained leading to formation of a different Delaunay tessellation. This results in a watermark of poor quality or even entirely lost in some instances. A method of i mproving the estimation of the nature of attack and thereby leading to an improvement in the quality of the extracted watermark is proposed in this thesis . The method is based on the estimation of the RST attack of an image by taking a mean distortion on s elected triangular regions. The accuracy of the proposed method has been tested by comput er simulation experiments using MATLAB . Various forms of attacks such as image filtering, image noising, histogram equalization, image cropping, JPEG compression and R ST attacks on standard test images have been simulated and watermark recovery assessed . The test images that have been include the 256 × 256 8 - bit grey level images ‘Pepper’, ‘Cameraman’ and ‘H ouse’ and the 512 × 512 8 - bit grey level images ‘ L ifting body’, ‘Gi rl’ and ‘ Baboon' among others. The proposed algorithm has also been tested on colour images such as ‘Baboon’, ‘Couple’ and ‘House’ and on video such as ‘Bus’, ‘Foreman’ and ‘Container’ among others. The P eak S ignal - to - N oise R atio (PSNR) and the Normalised Cross - correlation (NC) measures have been used to assess the quality of the watermarked images and recovered watermarks. The computer simulation results show that the proposed algorithm is effective in extraction of watermarks from images subjected to geom etric attacks.