Research On Payload Location Of Image Steganography

As one of important technical tools for defending information security, image steganography and steganalysis have become very attractive hotspots in the field of information security. Image steganography is a technique which first embeds the secret message into an innocuous image cover by virtue of the data redundancy and then transmits the image in public channel. Steganalysis, as the opposite technology against steganography, aims at detecting, extracting, restoring, and destroying the secret message embedded in the image cover.In recent years, detection techniques for image steganography have achieved fruitful results. Both specific and universal detection algorithms exhibit excellent performance under the laboratory environments. Some of them can even estimate the embedding rate accurately. However, the eventual destination of steganalysis is to extract and restore the secret message. Because of the absence of the location information about payloads, it is difficult to apply the results of detection and estimation of the embedding rate to the next logical step--extracting and restoring the secret message. As a result, it is significant to make a research on payload location. This dissertation firstly introduces the basic theories of information hiding and steganography, focusing on the status of payload location, and then based on existing knowledge, develops several payload location algorithms. The main contributions of this thesis are summarized as follows:1、Combining the existing Weighed-Stego(WS) residuals with Maximum a posteriori(MAP) estimator, an improved method for payload location is proposed. First, we use WS residuals to locate payloads originally, and then take the results as a priori information to apply to the modified MAP cover estimator to obtain a more accurate cover estimator. In the end, the residuals are computed to locate payloads finally. Experimental results demonstrate effectiveness of the new approach in locating payloads both on jpeg decompressed images and originally lossless stored images embedded with Least Significant Bit(LSB) replacement. Furthermore, the improved approach is suitable for payload location with different embedding rates well.2、A new method aimed at LSB replacement is proposed to locate payloads for stego images created by multi-keys(n). The existing methods for payload location are proposed with the same premise that the stego images are of the same size and have the same payloads, that is to say, the same key is used to create a large number of stego images. However, in practice, taking security into account, the key has to be changed termly. There will appear stego images created by multi-keys(n), which is very common. First, all payloads created by the n kinds of keys can be located based on WS residuals, then the cover estimation of every stego image is obtained with MAP estimator, the weighed graph can be acquired according to the modified pixels and the payload locations are separated to n different groups(corresponding to n different keys). In the end, the modified pixels of every stego image are compared to every kind of locations that have been classified according to different keys in order to identify the key that has been used to create the stego image, that is to say, payloads of every stego are located. Experimental results demonstrate the proposed method can not only separate all the payloads to every kind of key accurately, but also attach every stego image to the corresponding kind of key. In other words, payloads are located for every stego image.3、A new method is proposed aimed at LSB matching. The existing algorithms locate payloads by estimating the cover image and then computing the mean residuals. From a new perspective, the dissertation proposes a method of payload location aimed at LSB matching. Taking the problem of payload location as a matter of two-classification in pattern recognition, it can be solved by extracting the statistic feature of the square of adjacency pixel difference for every pixel and feeding into SVM classifier to classify all the pixels to two parts: payload or non-payload. The feature is proved effective theoretically. Furthermore, the proposed method is compared to the existing algorithms aimed at LSB matching. When the embedding rate is low, our method performs much better than the MAP estimator.Finally, the research work of this thesis is summarized and the further research topics and directions in the future concerning payload location are discussed.