Robust Image Watermarking Based On Geometry Invariance

Recently, digital images serve as a key media for information acquisition andtransmission and are therefore widely used in different fields due to some characteristics,such as direct viewing, vividness and cross-language. However, the fast development ofthe information technology has brought to forefront security concerns with the use ofdigital images. To solve the above problem, the digital watermarking technique has beenproposed and rapidly developed in the past twenty years. Usually, stego imagesinevitably suffer from the common image processing operations and geometric attacksduring the transmission, which is an important bottleneck to their usage in practicalapplications. Therefore, robust watermarking techniques for digital images haveattracted increasing attention for copyright protection in multimedia informationsecurity.To target the challenging problems of the robust image watermarking, this thesismakes an extensive study of robust image watermark based on geometry invariancefrom such aspects as geometry rectification, watermark embedding and detection modeland security analysis of watermark system. The main contents of this paper aresummarized as follows.(1) An image rectification method based on SIFT feature points matching isproposed. Firstly, the geometric distortion parameters of stego images are computed bymatching SIFT feature points from original images and stego versions. Following this,the invert transform is employed to restore image content. Based on these properties, theproposed method can accomplish geometry rectification for distorted stego images.(2) Since the rectification method based on feature matching can only handlelimited kinds of geometric attacks such as rotation and scaling transformations, a novelrectification method using RANSAC iteration algorithm is designed in this thesis.Following the feature matching process, the feature point pairs matched wrongly arefurther removed and the geometric transform model can be accurately determined byiteration procedure. In this way, the distorted images can be restored, which not onlyenhances the image quality but also improves the adaptibiliby of image geometricrectification.(3) By considering the effects of geometric attacks on stego images in practicalapplications, an image watermarking method based on feature matching in the Curvelet domain is proposed against geometric attacks, which combines the stable image featuresand embedding strategy in transform domain. The distorted image is firstly restoredusing feature matching in order to resynchronize the watermarks and images.Watermarks are embedded into the restored stego images by modifying Curveletcoefficients, which leads to an improvement of robustness and capacity of watermarks.(4) To further enhance the robustness against geometric attacks, an imagewatermarking method based on iteration optimization and Zernike moment is proposed.Firstly, both the graph-based segmentation algorithm and selection strategy of featureregions are used to obtain the stable and non-overlapping feature regions. Then, thewatermark embedding and detection algorithm based on the Zernike moment isdesigned. Finally, the geometry rectification model is optimized by the RANSACiteration algorithm, which is helpful for the improvement of imperceptibility androbustness of watermarks.(5) To further improve the security of the watermarking system, a multipleencryption algorithm based on cellular automata (CA) for robust watermarking isproposed. Firstly, the watermarked image is converted into binary sequences. Secondly,different CAs rules are used for encryption of side information, watermarks and stegoimages. Finally, the encrypted stego images can be obtained. The proposed methodeffectively solves the security problem of the watermarking system.Generally, we integrate the basic theories of pattern classification and computervision into robust image watermarking mechanism and propose five novel methods indifferent aspects. The proposed methods can overcome the drawbacks of the existingmethods and provide new ways for copyright protection and content authentication ofdigital images.