Research On High-dimensional Histogram Based Reversible Data Hiding For Images

Reversible data hiding technology for images embeds secret message invisibly into a digital image carrier,and the resulting marked image can be used normally.In specific scenarios,the hidden message can be accurately extracted from the marked image for the purpose of content security authentication,and the marked image can be completely recovered to its original state without any loss,eliminating the effect of data hiding on the distortion of the original image.This technology can achieve a role of pre-emptive protection and active defense of image content,and has important applications in sensitive fields that emphasize image content security,such as national security maintenance,judicial authenticity authentication,and medical image processing.In this dissertation,for different formats of images,three high fidelity reversible data hiding schemes are designed by exploiting the redundancy and high order statistical properties of images,in which the problems in the two stages of histogram generation and histogram modification are analyzed and studied.The main contributions are as follows.(1)We propose a non-local graph-based prediction method for reversible data hiding for gray-scale images.Formalizing pixel prediction as a graph signal restoration problem,a strategy for computing the weights of target pixel patch based on nonlocal neighborhood search is designed,and the pixel prediction values are solved by constructing a optimization problem using the quadratic graph smoothness prior and the graph total variation prior as signal priors,respectively.Further,a structure tensor based pixel ranking strategy is proposed to preferentially select smoother regions for embedding.Experimental results show that the proposed method performs better than two mainstream algorithms,achieving PSNR gains of 1.33 dB and 0.99 dB on average,respectively.(2)We propose an adaptive two-dimensional mapping and multiple histograms modification based reversible data hiding method for gray-scale images.According to the distribution of two-dimensional prediction error histogram,an adaptive adjustment method of two-dimensional mapping is designed to perform adaptive updating and iteration of the mapping in the direction of decreasing embedding distortion.To extend the traditional single histogram modification to the more complex multi-histogram case,a feasible and efficient solution is given for multi-parameter collaborative optimization problems.Experimental results show that the proposed method achieves better embedding performance than several state-of-the-art algorithms,where the PSNR is up to 64.33 dB on average for image dataset BOSSBase v1.01 at an embedding capacity of 10,000 bits.(3)We propose an adaptive three-dimensional mapping optimization based reversible data hiding method for color images.A three-dimensional prediction-error histogram is generated by cross-channel pairing,and a high-dimensional mapping adaptive optimization strategy is designed for this high-dimensional histogram.After ranking according to the frequency of the histogram,the three-dimensional mapping is adjusted and optimized in an ordered iterative manner,thus solving the complex optimization problem in high-dimensional space in a fast and lowcomplexity manner.Experimental results show that the embedding performance of the proposed method is better than two mainstream algorithms,where the PSNR is up to 61.77 dB on average for image dataset Kodak at an embedding capacity of 50,000 bits.