Methods for improved robustness of image watermarking algorithms

Abstract

With the advent of multimedia technology and the popularity of Internet communications, there has been great interest in using digital watermarks for the purpose of copy protection and content authentication. Digital watermarking technology allows the content owner to embed a secret signature, i.e., watermark, into the host content for many applications. For digital watermarking, the major challenge lies in the confident verification of the embedded watermark, even after the watermarked content undergoes various forms of unintentional or malicious modification. Approaches aiming to guarantee reliable verification of an imperceptible watermark are termed robust watermarking algorithms. In this dissertation, we study digital image watermarking and provide more robust algorithms toward reliable watermark verification, assuming various types of "content-preserving" image processing. Three new algorithms based on attack analysis, spectrum equalization, and a modified embedding rule are proposed. We discuss and analyze the proposed solutions, and compare them thoroughly against conventional algorithms. Since the watermark robustness is to be tested under various forms of image processing, the watermark encoder can utilize the knowledge of some possible attacks for a more secure embedding. Our first solution toward robust image watermarking is to select the set of best watermarking coefficients through attack analysis using the un-watermarked, original image. For transform-domain algorithms, the discrete cosine transform (DCT) or discrete wavelet transform (DWT) are normally used for decomposing the host image before embedding the watermark. Due to the low-pass characteristic of most images, the DCT/DWT coefficients generally vary in amplitude throughout the image spectrum. This low-pass nature is an advantage for many transform coders, but it does not facilitate a reliable watermark extraction for many watermarking algorithms. Our second solution for a more robust watermarking is the use of a simple, invertible permutation operator to equalize the transform coefficients before watermarking. Many transform-domain schemes utilize a directly-proportional rule for embedding the watermark. This approach results in diminishing performance as the watermark capacity increases. Our third solution provides a new embedding scheme that is inversely dependent on the magnitude of the selected transform coefficients. This scheme enhances performance, enabling a large-capacity watermark.