Zero distortion-based steganography for handwritten signature

The growth of the Internet over the last few years has enabled many people and organisations, such as financial institutions, around the world communicate with each other and transfer information over public channels. In this light, public channels are used due the lack of private network infrastructure and high setup cost of private networks. However, the data would be transferred through several different networks before being delivered to the recipient and the information can be read or modified by unauthorized user(s). To overcome this problem, steganography can be utilised as a solution for privacy problems in public networks, such as the Internet, where many digital media, such as images, audio and texts exist. Moreover, with the advancement of steganography, several researchers have recently devised steganalysis techniques, which threaten the steganographic systems. This means that any changes on the cover media (c) could lead to the identification of the stego media (s), which contains the secret message (m). Thus, developing a steganographic algorithm to use cover media (c) without raising attention is the most challenging task in data hiding. In this thesis, the human handwritten signature is introduced as a novel cover media (c) in conjunction with a steganography algorithm since there is a level of variability (i.e intra-user variability) within handwritten signature samples of an individual. To the best of our knowledge, this is the first time that a human handwritten signature sample is used for steganography application. In its simplest form, the existence of intra-user variability within handwritten signature samples of an individual is explored using the Kruskal-Wallis hypothesis test. Next, hiding data was accomplished by implementing a signature synthesis technique to produce a synthetic signature sample as a stego signature (s). This step was conducted by modelling both time series signals x and y (i.e. shape) of the handwritten signature samples using the maximum overlap discrete wavelet transform (MODWT) and several curve fitting techniques as the distortion function. Thus, the generated stego signature (s) is used to make stego key (k) based on the zero-distortion approach to represent the secret message (m) in a binary format. Finally, a computer numerical control (CNC) machine is utilized to plot the stego signature (s) on a piece of A4 paper for delivering to the recipient. On the other hand, by delivering the genuine signature sample as well as the stego key (k) using different channels such as the Internet, various image-processing techniques applied on the scanned stego signature (s) image to reconstruct the secret message (m). It was found that the acceptable range for the intra-user variability for genuine signature samples in the SIGMA signature database can be shown as Mean ± 2STD for both time series signals x and y. In addition, the imperceptibility rates of 3.5% and 4.7% were obtained for machine learning and human perception evaluation approaches, respectively, when identifying the stego signatures (s). This study has also demonstrated the payload capacity rate as 45.17%, which was the average percentage of usage of the stego signature (s) for encoding the predefined secret message (m). Finally, the proposed technique was able to retrieve the hidden data using the selected offline stego signature sample (s), with 94.7% accuracy rate.

Preview Text FK 2018 78 IR.pdf Download (969kB) | Preview

Actions (login required)

View Item