Model And Applications Of Reversible Data Hiding Under Inconsistent Distortion Metrics

With the popularity of electronic devices such as mobile phones and computers,digital media has shown explosive growth.In the Internet environment,access to docu-ments is becoming easier and easier.How to effectively protect copyright and integrity authentication of digital media becomes very important.Adding watermarking into media is an effective method for data copyright protection and integrity authentication.Watermarking embedding has a slight impact on the host file,which is imperceptible to the human eye,but the damage is permanent.For some precious files,such as med-ical,military and judicial files,minor modifications are not allowed.In this case,after embedding the watermarking into the host,it is necessary to be able to restore the host losslessly after extracting the watermarking,thus reversible data hiding(RDH)emerges as the tirmes require.RDH embeds watermarking into the host file to generate the stego file,from which the receiver can losslessly recover the host file and the watermarking.At present,most RDH algorithms define consistent distortion metric for host se-quences.Taking image as example,the cost of each pixel's modification is independent of its location but only related to the modification magnitude.However,from the per-spective of security,the modification of complex texture areas of images is not easy to cause the attention,which is more suitable for hiding information than the smoothing areas,thus the modification weights in smoothing areas and complex areas should ob-viously be different.Therefore,it is more reasonable to define inconsistent distortion metrics for host sequences,which are related to both the modification magnitudes and locations.Based on inconsistent distortion metrics,this dissertation establishes the corre-sponding theoretical model,and successfully applies such model to the research of RDH in color image,covert storage and reversible image processing.The main research works and innovations of this dissertation can be summarized as follows:1.Proposed Theoretical Model of RDH under Inconsistent Distortion MetricsCurrent RDH optimal coding is only suitable for position-independent consis-tent distortion metric,while position-dependent inconsistent distortion metrics are more reasonable in applications.In this dissertation,inconsistent distortion metrics are quantized into a multi-distortion metric,and the rate-distortion bound under multi-distortion metric is presented.Multi-distortion metric model can be transformed into consistent metric model,and then be solved by using the exist-ing methods for consistent metric model.Both theory and experiment prove that the proposed algorithm makes RDH optimal coding be optimized under multi-distortion metric,which greatly expands the application of optimal coding.2.Proposed Two Algorithms for RDH in Color Image under Inconsistent Dis-tortion MetricsMost of RDH algorithms in color image do not take into account the distinctions of human eyes for different colors,while for color images different distortion met-rics should be defined for different color channels.In this dissertation,adaptive distortion metrics are defined for the three channels of red,green and blue re-spectively.By applying the RDH theoretical model under inconsistent distortion metrics,the problem of modification optimization for three channels is success-fully solved,thus minimizing the weighted distortion.Experimental results show that the visual quality of the stego image generated by the proposed scheme can be greatly improved.On the other hand,current RDH algorithms modify the host image to generate the stego image,which introduces modification distortion,and distortion may cause the post-processing of the stego image,such as classification recognition,to be in-terfered.Many color image processing algorithms such as SIFT(Scale Invariant Feature Transform)firstly converts the color image into a grayscale image,and then performs operation on the grayscale image.In this dissertation,by keeping the gray value of the color image unchanged,some post processing of color image is not disturbed.Experimental results show that the stego image generated by the proposed algorithm does not affect the accuracy of image processing algorithms based on grayscale image such as SIFT feature matching.3.Proposed Two Types of Algorithms for Covert Storage Based on RDH under Inconsistent MetricsCurrent encryption-based secure storage methods store files in meaningless messy codes,while tends to expose the file to be a confidential file and be subject to at-tack.Covert storage is distinguished from traditional storage methods,which hides secret files in natural files such as natural images,making secret files un-detectable.Considering the erasable property of covert storage,it is natural to associate with RDH.Traditional RDH algorithms mainly focus the modification on the smoothing area of image,and is easy to be detected thus unsafe.In this dis-sertation,the position-dependent adaptive steganography distortion metrics are defined for the host sequence,and then the optimal transition probability of the host to the stego is calculated.RDH with anti-detection capability can be realized by the optimal coding according to the transition probability,called reversible steganography.Experiments show that the proposed reversible steganography algorithm has higher anti-detection capability.The above reversible steganography is suitable for embedding a small amount of information into a large host file.For the case where the secret file is an image,we have proposed two reversible visual transformation techniques.Reversible visual transformation can be regarded as a large-capacity data hiding technology for images,which converts a secret image into an unrelated target image to generate a camouflage image.Experiments show that the visual quality of the camouflage image generated by the proposed scheme is greatly improved,which is highly similar to the target image thus provides camouflage for the secret image,and the secret image can be restored without loss from the camouflage image.4.Proposed Reversible Image Processing Based on RDH under Inconsistent MetricsCurrent image processing algorithm is irreversible,that is,the original image is damaged after processing,while reversible image processing will bring a lot of convenience.For many image processing algorithms such as contrast enhance-ment,gamma transformation,etc.,after processing the correlations between the obtained target image and the original image are very high,thus the original im-age can be effectively compressed based on the target image.This dissertation proposes the following framework for reversible image processing.The client uses any image processing software or algorithm to process the original image to obtain the target image,then compresses the original image according to the target image,and finally reversibly embeds the compressed secret image into tar-get image through RDH to obtain the reversible processed image.Experiments show that as for contrast enhancement,gamma transformation,etc.,reversible processed image are almost the same as the target image visually,but from which the original image can be restored.