Research On Image Reversible Data Hiding Based On Pixel Correlations

With the rapid development of Internet technology and cloud computing technology,the issue of information security is becoming increasingly prominent.As one of the important means of protecting information security,data hiding is the embedding of additional information into a carrier without affecting the use of the carrier in order to achieve functions such as copyright protection,covert communication and content integrity authentication.Reversible data hiding allows authorised users to restore the original carrier without damage after the extra information has been correctly extracted.Due to this property,reversible data hiding techniques have received widespread attention in some areas such as military,medical and legal fields that are sensitive to changes in the original image content.This thesis investigates reversible data hiding algorithms for images based on pixel correlations.The pixel correlations of original image are used to investigate high-fidelity plaintext image reversible data hiding algorithm to enhance the visual quality of images,thereby improving the security of covert communications,and high-capacity reversible information hiding in encrypted image(RDHEI)based on reserving room before encryption(RRBE).By preserving the pixel correlations of original image,reversible data hiding method based on vacating room after encryption(VRAE)and reversible data hiding in encrypted images based on image sharing are investigated to improve the hiding capacity and robustness.The main research work of this thesis is as follows.(1)A reversible information hiding algorithm based on pairwise prediction error expansion(PEE)and two-dimensional histogram decomposition is proposed.The pixel-based pixel value ordering(PPVO)algorithm hides information in onedimensional space using only pixels whose values are outside the range of the minimum and maximum values of the neighbouring pixels.To address this problem,this algorithm uses pixels whose values are within the range of the minimum and maximum values of the neighbourhood pixels to hide information by considering the correlation between the current pixel and its neighbourhood pixels so that the number of pixels that can be hidden is increased and thus the hiding capacity is increased.Based on the correlation of neighbourhood prediction differences,the one-dimensional PPVO hiding is extended to twodimensional space.Prediction error pairs are counted to obtain a global two-dimensional prediction error histogram.The global 2-D prediction error histogram into is decompose multiple sub 2-D histograms based on the relationship between the difference between the maximum and minimum values of neighbouring pixels and the current pixel.For the different sub-2D histograms,the two-dimensional mapping method is chosen to hide the information and modification distortion can be reduced according to the characteristics of histogram distribution.The experimental results show that the image quality the proposed method is better than some state-of-the-art schemes for the same embedding capacity.(2)A reversible data hiding algorithm for encrypted images based on hierarchical embedding is proposed.Most reversible data hiding algorithms for encrypted images do not make full use of the pixel correlations of the plaintext image before encryption,resulting in fewer bit planes for hiding,which in turn leads to low hiding capacity.To address the above issue,this algorithm generates prediction error values based on the correlation between the pixels of the original image before encryption.A hierarchical trinomial tree model is developed to divide the absolute value of the prediction error into small,medium,and large ranges of value and to generate a bit-plane label.In each layer of the trinomial tree,the bit planes corresponding to the small and large ranges are used to hide information,thus increasing the embedding capacity.The original image is encrypted using stream cipher.During data hiding,the bitplane label map is compressed and embedded into the bit-plane where the encrypted image can be hidden,and the remaining concealable bit-plane is used to embed secret information.When extracting,the bit-plane label map is first extracted,and the unique combination of bit-planes is solved according to the label map to obtain the hidden information and recover the original image.The experimental results show that the algorithm has a large embedding capacity and outperforms many similar algorithms.(3)A reversible data hiding algorithm for encrypted images based on adaptive difference recovery is proposedThe reversible data hiding algorithm based on vacating room after encryption(VRAE)has a low capacity.In this thesis,we first design an adaptive difference recovery technique(ADR),which determines the original pixel difference range according to the hidden pixel difference range,and then uses dichotomous search to recover the original pixels efficiently and losslessly by combining the original pixel difference and the reference pixels.Block permutation and block based modulation are used to encrypt the original image,so that the pixels within the encrypted image block retain the pixel correlation within the original image block.The ADR technique is applied to the encrypted image to achieve RDHEI with high embedding capacity.The algorithm hides information directly in the encrypted image without pre-processing operations.Three datasets,BOWS-2,BOSSbase and UCID,are used as carriers to obtain the average embedding capacity of 3.1145 bpp,3.2045 bpp and 2.4633 bpp,respectively,which achieve high payload RDHEI based on the VRAE framework.The experimental results show that the algorithm has better embedding performance compared with the mainstream algorithms.(4)A reversible data hiding algorithm for encrypted domains based on image sharing is proposed.Aiming at the poor anti-attack capability of most current RDHEI algorithms,and the problems of pre-processing,data expansion and low capacity of RDHEI with secret sharing,this thesis proposes an encrypted domain reversible data hiding algorithm based on image sharing.In the image encryption phase,a new iterative image encryption method is designed,which makes the encrypted block retain the pixel correlation within the original image block pecfectly.In the image sharing phase,an image sharing scheme with imposed constraints is used to generate multiple encrypted shares such that the blocks of each encrypted share preserve the pixel correlation of the encrypted blocks well.The above two stages allow the blocks of each encrypted share to retain the pixel correlation of the original image blocks well without causing data expansion in the encrypted share.When hidden,chunking is performed for each encrypted share.Block flag information is generated based on the pixel correlation within the block and information is embedded within the chunks to achieve high payload.At the receiving size,even if part of the encrypted share is corrupted or lost,the original image can be recovered if enough complete encrypted shares are received,improving the robustness of reversible information hiding in the encrypted domain.