Abstract:With the development and maturity of digital information and cloud computing technology, reversible data hiding in encrypted images are gradually becoming a research hotspot for data privacy protection in communication transmission. It is still a big challenge to realize more complete compression of carriers to enhance embedding capacity by adaptive recognition coding of pixels in local areas based on the distribution characteristics of pixels in different cover images. To address the problem of low embedding capacity of reversible information hiding in encrypted image due to insufficient utilization of redundant space in carrier images, a cipher-text domain reversible data hiding algorithm based on difference coding and block compression is proposed. Firstly, the difference image matrix is classified and adaptively coded according to the maximum difference value within the block based on the difference of strong correlation neighbors in the local area of the natural image, At the same time, the original image is chunked, disrupting the order between blocks and diffusion of pixels within blocks to ensure the image information security. Then, according to the result of the adaptive coding of the difference pixel blocks, the redundant space is compressed on the corresponding encrypted image blocks, and finally the secret data is embedded by bit replacement. Different from the previous algorithm, this paper rearranges the bits of the differential-encoded image, starting from the last significant bit of the cipher-text image pixel, marking the sign bit of the differential pixel, and marking the least differential bit in each of the remaining planes in turn. Due to the reversibility of the encryption operation and adaptive coding, the legitimate receiver can achieve lossless reconstruction of the original plain-text image and error-free extraction of the secret data. The experimental results show that the proposed algorithm has higher embedding rate and better security compared with several existing algorithms, and the average embedding rate reaches 3.027bpp and 2.937bpp on both BOSSbase and BOWS-2 datasets, and the average embedding rate is also improved by more than 0.57bpp on classical test images.