Digital watermarking techniques for image data

Abstract

In recent years, using digital watermarking for copyright protection of multimedia data has been actively investigated by people from different communities. Digital watermarking is a technique that inserts some information such as text, number or image into multimedia data through slight modification of the data. The work described in this thesis is concerned with the design of a high security watermark embedding and extraction algorithm and the design of a robust watermarking algorithm against geometric distortion. The security issues of different watermark embedding and extraction methods are firstly investigated. In most of the existing schemes, we find that the original embedded content can be easily extracted even if we do not have the correct key. Therefore, the attackers can easily guess the original watermark correctly by the extracted content if no extra operation such as bit relocation is applied before embedding. To solve this problem, we have developed a new scalar quantization based method by increasing the total number of possible lookup tables. The results show that our algorithm has a better security performance than the traditional scalar quantization based method. For geometric distortion resilient watermarking, we find that it is difficult to estimate the original signal without the original image in the frequency domain when the marked image has geometric distortion. In the spatial domain, in some cases, we find that the feature points of an image may give incorrect information to the watermarking algorithm such that the robustness is decreased. In order to improve the robustness, we have developed an algorithm to generate a hexagonal watermark pattern. The results show that our pattern can improve the robustness of a wavelet based watermarking algorithm against image cropping and has better performance for content authentication applications. In addition, me have developed a new DCT based embedding algorithm against simple geometric distortion by making use of the properties of DCT coefficients. Compared with the traditional DCT based embedding method, our method can give an error free extracted watermark even if the marked image is rotated at particular degrees or flipped.

Date of Award	16 Feb 2004
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Jian FENG (Supervisor)