Payload Location For JPEG Image Steganography Based On Co-frequency Sub-image

Currently,JPEG images are most widely used in people's daily work and life among images in many formats.Correspondingly,in the past ten years,JPEG image steganography and steganographic forensics have been one of the hotspots in the field of information hiding.However,the current research on JPEG image steganographic forensics mainly focuses on the detection of stego images,which is still a certain distance away from the ultimate goal—extracting the embedded messages.The existing research shows that the high accuracy payload location results will effectively improve the success rate and efficiency to extract the embedded messages.From this,on the basis of systematically summarizing state of the art of payload location,this thesis focuses on the payload location for JPEG image steganography.The main contributions of this thesis are summarized as follows:1.A payload location framework based on the co-frequency sub-image is proposed for JPEG image steganography.On the basis of summarizing the position selection mechanisms in JPEG image steganography into all coefficients pseudo-random scrambling method and embeddable coefficients pseudo-random scrambling method,the basic principle of payload location for JPEG image steganography is analyzed,and it is pointed out that the key of payload location lies in the accuracy of cover estimation.Then,a cover JPEG image estimation method is proposed based on co-frequency sub-image.The proposed framework computes the residuals of quantized DCT coefficients from the given multiple stego images and the cover images estimated by above cover estimation method,then fuses them to locate the embedded messages.2.Some payload location algorithms based on wavelet filtering the co-frequency sub-image are proposed.Under the condition of owning multiple stego images embedded along the same embedding path,the proposed algorithms firstly perform low-pass wavelet filtering on each of the co-frequency sub-images to estimate the cover co-frequency sub-images,then combine the estimated cover co-frequency sub-images into the estimated cover JPEG image.Based on the characteristics of the steganography change and whether the coefficients are embeddable,calculate the quantized DCT coefficient residual between the coefficients in the estimated cover image and the stego image.Then fuse the residuals at the same position of multiple stego images by mean to judge whether the position is stego position.Experimental results show that the proposed algorithms can accurately locate the stego positions of typical JPEG images steganography,such as JSteg and F5 steganography,and even if the investigator owns only 10 stego images embedded along the same embedding path,the payload location results still can be used to effectively recover the stego key,then extract the embedded messages.3.A payload location algorithm based on optimal estimation of cover co-frequency sub-image is proposed.Under the condition that there are multiple stego images embedded along the same embedding path,the proposed algorithm firstly uses the first-order Markov model to model the co-frequency sub-images,and uses the Viterbi algorithm to compute the optimal estimations of cover co-frequency sub-image for four scanning modes.Then for each scanning mode,the obtained optimal estimations of cover co-frequency sub-images are combined into an estimated cover JPEG image.Then,the absolute differences between the embeddable coefficients in the same position of the stego image and each of the four estimated cover images of four scan modes are computed and used as the residuals.Finally,fuse all residuals in the same position of multiple stego images by mean to judge whether the position is stego position.Experimental results show that the proposed algorithm can recognize the stego positions of F5 steganography in low frequency with higher accuracy then the payload location algorithm based on co-frequency subimage wavelet filtering,and can make up for the low accuracy of it in the low frequency.4.A payload location algorithm based on the fusion of co-frequency sub-image Gabor residuals is proposed.Because the 2D Gabor filters can capture the correlation in images in different scales and different directions,under the condition of owning only a single stego image,the proposed algorithm firstly uses multiple 2D Gabor filters with different scales,directions and phase parameters to filter the co-frequency sub-image to obtain multiple residuals at each position,then uses the ensemble classifier to fuse multiple residuals at the same position to judge whether the position has been modified.Experimental results show that the proposed algorithm can locate the modified coefficients with high accuracy even under the condition of only a single stego image,and even if the modification ratio is as low as 0.1 and the location accuracy of other payload location algorithms is lower than 10%,the proposed algorithm based on co-frequency sub-image Gabor residual can still locate the modified coefficients with accuracy close to 80%.Finally,a conclusion is given and the future researches are discussed.