Research On Image Authentication Technology Of Invariant Feature-based Digital Watermarking And Perceptual Hash

The amount of data has been exploding in the era of big data. Data have swept intoalmost every aspect of our daily lives from business to leisure and even society. Data are nowthe essential factor of production that much of modern innovation and growth simply can nottake place without them. With the torrent of data flooding our world, the coming ofinformation security risk is not only becoming more and more complicated with increasingharm, but also invoving more deeply to harm people’s privacy.As one of the important multimedia for information propagation, image is widely used inmany applications，and is facing three main aspects of the safety problems under the era ofbig data. Firstly, image is easier to be reproduced, tampered and illegal dissemination on theInternet. Secondly, the lack of planning and managing of massive digital images is easy tocause the intellectual property disputes, and damages the legitimate interests of the contentprovider. Thirdly, the lack of effective supervision of the internet pornography, violence, andprivacy image resources seriously affectes the healthy growth of the teenagers and destroysthe harmonious development of society. Thus, how to certification and recognition of imagecontents and how to protect the authenticity and completeness of the image contents havebecome an urgent problem, which has attracted widerspread concern in society.Unlike the traditional cryptography technique, digital watermarking and perceptualhashing technology are no longer passive intrusion prevention, but active intrusion preventionsecurity mechanism. They are the two most promising techniques to solve the above threeproblems, and image authentication methods based on digital watermarking and perceptualhashing technology have become a hot research topic in the field of multimedia informationsecurity.This thesis begins with an overview of digital image authentication technology, includingthe related theories and classical algorithms. For many complex issues in the practicalapplications, we proposed three digital watermarking authentication schemes based oninvariant image features, and designed two image hash authentication schemes based on theshape invariant moments and quaternion theory respectively. Contributes of this thesis are asfollows: (1) A reliable digital image watermarking algorithm based on singular valuedecomposition (SVD) is proposed to solve the false alarm problem of the traditional SVDalgorithm. The main idea is behind the using of the similarity measure of orthogonal Ucomponent. Moreover a meta-heuristic artificial bee colony algorithm is used to optimize thewatermark scaling factor, making it achieve a good balance between robustness andimperceptibility.(2) A robust visual cryptography copyright protection scheme is developed. Firstly,digital wavelet decomposition (DWT) is applied to decompose the cover image, and artificialbee colony algorithm is used to adaptive select feature sub-blocks. Then SVD is implementedto extract the image feature, which is classified into two clusters for generating the mastershare. Finally, ownership share is generated with the help of the secret image (watermark) andthe master share. In the process of ownership authentication, the secret image can be acquiredby stacking the master share and the ownership share. Experimental results show that theproposed scheme is robustness and can effectively resist several different signal processingattacks, without any distortion on the cover image.(3) Effective perceptual hashing algorithm based on radial Tchebichef momentsinvariants is proposed. The radial Tchebichef moments constituted by Tchebichef orthogonalkernel pose good orthogonality and robustness. We therefore apply these radial Tchebichefmoments to extract the image luminance information, and use the adaptive quantizationmethod to generate the hash feature vectors. The hash feature vectors is mapping to the finalhash after random Gray code binarization and chaos scrambling. Experimental resultsdemonstrate the good robustness and discrimination of the proposed algorithm.(4) Since conventional image hash methods to deal with color images are simply basedon luminance components, which may lose some significant color information, and havelimited discriminative capacities，we therefore propose a robust hash algorithm to deal withcolor images based on the quaternion polar harmonic transform (QPHT) moments. The ideaof the algorithm consists of representing three channels of color images as a pure quaternionand exploiting the invariance property of the QPHT moments to extract robust hashes. Thisnew method not only makes up the deficiency of the traditional color features in describingthe specific shape characteristics of the object, but also solves the insensitive to color change of the traditional shape invariant features.(5) We have proposed a robust image regularization based algorithm for image protection.The algorithm combines the advantages of SVD, Fuzzy-Possibilistic C-Means (FPCM) andimage regularization, where the image regularization enhances ability to resist geometricattacks, SVD and FPCM are used to generate the verification map efficiently. With thisverification map, anyone can verify the image copyright practically without the help of theoriginal image. In addition, watermark chaotic scrambling by logistic chaotic map alsoimproves the security performance of the algorithm.