Research On Information Hiding Techniques Of Resisting Geometrical Attacks

With the development of multimedia processing technology and Internet, information hiding has become a hot interest area recently in the research of information security.Geometrical distortions can desynchronize the data-hiding scheme while the visual quality of image is preserved well.To design a robust and secure data hiding scheme resisting geometrical attacks is regarded as one of the most difficult and challenging work,which has become the hot point of research in recent years.This thesis is focused on the synchronization problems of data hiding scheme. Based on the in-depth analysis of feature point extraction of image in pattern recognition and fast correlation attacks in cryptanalysis,two novel data hiding algorithms are proposed for watermarking and secret communication with high capacity.The main contributions of this thesis are summarized as follows:1) An effective image feature points detection technique based on scale space theory is proposed.Feature points detection is the classical research topic in pattern recognition,which is also the basis of the second-generation watermarking scheme.With the scale space theory,we design a more robust Harris detector.2) Robustness to geometrical attacks is of importance for watermarking.Most watermarking algorithms proposed mainly address the problem of global affine transforms,such as RST,but the local geometrical attack is still intractable.In this paper we propose a novel local geometrical invariant watermark scheme based on fast correlation attacks of stream ciphers.The paper also explicitly analyses the channel noise model so as to provide a basis on which some important parameters used in fast correlation attacks are designed.Simulation results show that the proposed algorithm is robust against various attacks including RST,cropping,removing lines or columns, even RBA in StirMark and common signal processing such as JPEG compression,filter.3) We propose a novel geometrical invariant image data-hiding scheme that is composed of a synchronization technique by using soft fast correlation attacks(SFCA) and a DCT payload embedding method.The position of the center of gravity of each Delaunay triangle,as the synchronization information,is extended into the linear sequence that is additively embedded in the red channel of each corresponding triangle.In the step of the extraction,the synchronization information can be reconstructed robustly by applying the technique of SFCA used commonly in cryptanalysis and afterward be used to recover the affine transformation that has been done on the image.After this step,the secret messages embedded into the DCT domain of the blue channel of the image can be extracted.Simulation results show that the data-hiding scheme with high capacity is highly robust to geometrical distortions including RST,ARC and other affine transformations.In addition,it is also robust against common image processing such as JPEG compression and filters.4) Semi-fragile watermarking plays an important role in multimedia content authentication.Most previous techniques only focus on distinguishing the JPEG compression from the malicious attacks,and most of them reject the geometrical modification.However,in fact,in many content authentication applications geometrical distortion should be considered as the non-malicious manipulations.So,in this paper we propose a novel semi-fragile watermarking scheme that can solve such problem.The scheme can distinguish attacks according to the watermark generated from the semi-fragility of the moment invariants of the image.Another watermark produced from the approximate coefficients of the image's wavelet decomposition is utilized to localize the tempered regions.Simulation results show that the proposed scheme can correctly distinguish the malicious attacks from the incidental distortions and localize tempered regions accurately,and resist the forge attacks simultaneously.