A New Statistics Preserved JPEG-PM1 Steganography Algorithm

In this thesis, we firstly propose two modified Plus-Minus 1 algorithm in JPEG(J-PM1): a) the Plus-Minus 1 algorithm based on Least Square Matching (LS-JPM1), and b) the Plus-Minus 1 algorithm based on partial zero embedding (PZ-JPM1), each aiming to improve the security and capacity respectively. Secondly, we introduce the Shuffled Frog-Leaping algorithm(SFLA) in PZ-JPM1 to determine the optimal embedding parameters which better preserve the second order statistics. Finally, instead of the PM1 steganography, we modify Differential Energy Modulation (DEM) scheme for robust embedding. The main work includes:1. LS-JPM1 is proposed to maintain the original distribution of coefficients. According to the first order statistics of a host signal, the least square matching(LSM) is used to adapt the flipping probability of each coefficient to that of the host signal. LS-JPM1 obtains average 58.8% decrease of histogram distortion than J-PM1.2. Based on LS-JPM1, PZ-JPM1 is proposed to increase the capacity. Unlike traditional J-PM1 schemes, secret bits also alter some zero coefficients in PZ-JPM1. The number of zero-embedding coefficients and the latter's flipping probability are decided also by LSM method with respect to the host statistics signal. Compared to LS-JPM1, PZ-JPM1 gets average 29% increase of capacity.3. SFLA is applied to preserve the second order statistics. According to the flipping probability of each coefficient decided by PZ-JPM1, the SFLA is utilized to optimize the specific embedding positions to improve the capacity of preserving the stego image's second order statistics.4. DEM, an area embedding scheme, is modified to get better robustness and security. An updating energy distribution table(EDT) records an accumulating distortion and leads to the least distorted embedding. Average 77.1% decrease of histogram distortion and 2.01dB increase of imperceptibility than DEM are obtained.