Abstract
As the Internet becomes more and more populous, people concern more about the copyright protection issue for digital data such as images and audio. Digital watermarking technique can hide data in images or audio to indicate the data owner or recipient. Therefore, it can protect the copyright. Motivated by copyright protection in the Internet, we propose an Internet Image Library (IIL) using watermarks to protect the copyright. With this watermarking application– IIL in mind, we analyze and propose new watermark systems to meet this application. A lot of researchers discussed how to embed a watermark into image in recent years. However, few researches analyzed the watermark techniques in a theoretical point of view. In this thesis, we first interpret the watermarking problem as a digital communication problem. There are three main criteria concerning the performance of a watermarking technique– capacity, imperceptibility, and robustness. We then show the trade-offs between these three criteria adopting concepts from the digital communication theory. After the analysis has been established, two watermarking schemes adaptive to Human Visual System are proposed. In addition, error correction coding and spread spectrum communication technique are used to lower the watermark detection error. Consider the original image as the transmission channels, capacity is interpreted as the channel capacity of the image. Imperceptibility of the watermark is the signal to noise ratio (SNR) with watermark as the signal and image processing errors or/and original image as noise. Probability of watermark detection error demonstrates the robustness of the watermark technique. We analyze watermark systems by two approaches: channel capacity analysis and probability of watermark detection error analysis. The first analysis shows that we cannot increase the capacity without making the watermark obtrusive in viewing with the noise tolerance of the watermark system controlled. The watermark system is more robust when the watermark signal power rises with the capacity of the system controlled. In the second analysis, we know that in addition to stronger watermark power, we can use error correction codes and spread spectrum communication technique to lower the watermark detection error. Unlike the general digital communication system, where the channel characteristics is known or estimated, an image channel is hard to define because it can be transformed to many different domains, such as DCT domain, Wavelet-Transform domain, etc. Therefore, an image channel has different capacity and probability of detection error respect to different representations. In this thesis, we develop our system on two most commonly used domains, spatial and DCT domain. By exploiting the HVS characteristics, we develop two adaptive watermarking systems: Direct Sequence Spread Spectrum (DSSS) watermarking system based on Human Visual System (HVS) in spatial domain and Frequency Hopping Spread Spectrum watermarking system based on Human Visual System (HVS) in DCT domain. HVS is adopted to determine the watermark power constraint. Capacities of an image are then computed. Locations with larger capacity values are selected to embed the watermark. Watermark is derived from message been error correction coded and weighted adaptive to HVS values. After the analysis on watermarking systems and proposition of new watermarking techniques have been performed, a digital internet image library in the World Wide Web, call Internet Image Library (IIL), is presented as a watermark application for copyright protection in the Internet. The central concept in this system is to use watermarks embedded in the image to hide the creator, IIL, and the consumer information. The creator watermark stays permanently in the image. A suspected image is inspected if a rightful consumer watermark matches with the consumer personal information. The retrieved IIL watermark has add-on values such as leading consumer back to the IIL WWW site for further transaction or advertisements.