Reversible Data Hiding in Encrypted Images with Distributed Source Encoding: Implementation and Experiments

In this thesis, we implemented and investigated Qian-Zhang reversible data hiding scheme proposed in 2016. Qian-Zhang scheme uses Slepian-Wolf encoding based on Low-Density Parity-Check (LDPC) codes to compress selected most significant bits (MSB) from an encrypted image to vacate room for embedding additional data. Compressing process depends on LDPC matrix, H, r<n, where r is number of rows and n is number of columns. After extracting embedded data, the original image can be recovered by applying iterative decoding algorithm. We found that the quality of the recovered image depends on the construction method, size, and ratio R=r/n. We implemented Qian-Zhang scheme using H matrices constructed by two methods, Gallager and MacKay-Neal, having different sizes and ratios. We evaluated QianZhang scheme with these matrices using decoding time, embedding capacity, and quality of the recovered image, approximate and decoded, by Peak Signal-to-Noise Ratio (PSNR). We get a formula for embedding capacity dependence on the number of bits to be compressed and value of R. In addition, we investigated relation between PSNR of an approximate image and embedding capacity. Changing of the embedding capacity does not affect PSNR of the approximate image. Since we used other H matrices than the one used by Qian-Zhang, we obtained not exactly same PSNR and embedding capacity but close to the values of Qian-Zhang. In addition, we investigated the PSNR of decoded image when decoding fails. The PSNR decreases when the embedding capacity increases. We found that fixing ratio, R, and increasing size of H leads to the increase of the PSNR of the recovered image. On the other hand, the time of decoding increases with the matrix size growth. These results may be used for choosing suitable H matrix size to meet specified decoding time. We investigated relation between the ratio, R , and embedding capacity. Decreasing of R leads to the increase of the embedding capacity. We investigated relation between R and PSNR of the decoded image. Decreasing of R leads to the decrease of the PSNR. Our results show better embedding capacity than that in the Qian-Zhang’s paper due to the use of different size H matrices. Keywords: Reversible data hiding, Slepian-Wolf encoding, Low-Density ParityCheck (LDPC) code, LDPC matrix , Most Significant Bit (MSB), Distributed Source Decoding (DSD), Selection ratio, Embedding capacity, Host image, Approximate image, Decoded image, Peak Signal-to-Noise Ratio (PSNR).