Research On Reversible Data Hiding In Encrypted Images

Traditional Data Hiding(DH)technology only focuses on the security and invisibility of secret Data.Reversible Data Hiding(RDH)technology can not only hide the secret data but also recover the carrier image lossless,but it ignores the protection of the carrier image.As a branch of the Data Hiding field,Reversible Data Hiding in Encrypted Image(RDHEI)is used not only to protect the secret data but also to encrypt the carrier image to realize the purpose of protecting the carrier image.More and more scholars focus on reversible data hiding in encrypted images and propose corresponding algorithms.Considering the inherent properties of images,their high bit-plane data have better correlation than low bit-plane data,and it is also easier to get redundant space by compressed coding,so many scholars focus on the high bit-plane data of images.These methods perform very well in high bit-plane data compression,but for low bit-plane data compression efficiency is greatly reduced,and even in some cases extra space is needed to represent low bit-plane,resulting in poor data embedding rate and also security risks.Aiming at the problems of low embedding rate and low security existing in reversible data hiding algorithms in encrypted images,this thesis carries out relevant research around these problems,focusing on the traditional spatial methods and image feature expression methods.In the spatial algorithm,the difference coding is used to improve the embedding rate.To take into account the embedding capacity and security of the algorithm,sparse representation was used to reconstruct the image,and the adaptive threshold was used to classify and encode the image blocks.To meet the requirement of higher capacity,Auto Encoder is introduced for feature extraction and the secret data is embedded into the feature map.The main work and innovation of this thesis are as follows:(1)A method of reversible data hiding in encrypted images based on pixel difference coding is proposed.The image bit plane is no longer divided,the pixel is taken as a whole,and the similarity between pixels is fully utilized to obtain more redundant space.The MED(Median-Edge-Detector,MED)preprocessing was used to improve the redundancy of the image,and the histogram shifting was used to deal with the MED prediction error and the overflow pixel.In the stage of image encryption,the carrier image is divided into blocks and the encrypted carrier image is obtained by block scrambling and pixel scrambling.In the embedding stage,the image blocks are divided into 8 categories according to the maximum difference within the block,and the image blocks are encoded and compressed according to different types.A large number of experimental results show that this algorithm greatly improves the image payload.(2)A method of reversible data hiding in encrypted images based on block classification coding of sparse representation.In order to be applied in some scenes that require higher security of embedding,sparse representation is applied in encrypted images,block scrambling is adopted in the image encryption stage,and the correlation between image pixels is maintained at the image block level.Combined with the code length of error coding,different embedding methods are adopted for different image blocks according to the adaptive threshold,which not only improves the embedding rate of the secret data,but also improves the complexity of the algorithm to a certain extent,that is,improves the embedding security,and synchronously imparts the errors of sparse expression,so that lossless carrier images can be obtained in the extraction stage.(3)A method of reversible data hiding in encrypted images based on Auto-Encoder is proposed.In some application scenarios,the restoration quality of the carrier image is not high,but the embedded data capacity is higher.In order to obtain a higher embedding rate,Auto-Encoder is introduced in this thesis to extract the features of the encrypted image through the neural network.Block scrambling is used in the image encryption stage to maintain the correlation between image pixels at the image block level.In order to improve the quality of AE compression,multiple attempts are carried out on the AE network structure,and image data loss during neural network pooling is reduced by multi-route pooling and jump links.In some special application scenarios,in order to improve the quality of the recovered image,U-net is applied to the reversible data hiding in encrypted images,which greatly improves the quality of the recovered image by increasing the communication overhead.