Data hiding in still images

Abstract

In the era of easy communications, data is widely transfered through Internet and networks. A lot of devices are nowadays in use, like smart phones, tablets and netbooks, connected to the web by mobile Internet providers, wi-fi or hot spots, enabling the direct exchange of data and multimedia. This continuous transferring of information which has became an essential need of our everyday life, also generates risks that have not been observed before. Social networks connect people not only through instant messages, but also through unlimited multimedia. Here comes the need of proof of multimedia ownership, tamper detection, and/or secret communications. Data hiding finds applications in this area. A great effort has been made to hide information with no perceptual degradation, for use in copyright detection, where robustness is the basic requirement. The topic of this thesis is the development of data hiding techniques in images, where in contrary to what has been mentioned before, robustness is not demanded. The algorithms generated in this study that embed information in images for authentication purposes are fragile. The presence of the inserted message allows the verification of image authenticity and, eventually, the localization of the tampered regions. Design of fragile techniques has received the most attention of the research in this thesis. Applications of discrete Pascal and Gould transforms on image authentication have been implemented, extensively studied, and experiments have given very good results. For the further investigation of fragile data hiding methods texture classification patterns are studied. Local Binary Pattern and Rotation invariant Local Binary Pattern (LBPROT) have been proved to be very promising for information hiding in terms of capacity and image quality. Two reversible fragile data hiding methods are also presented in this thesis. The one is an improvement of difference expansion by using overlapping pairs of pixels, instead of non overlapping pairs. This increases capacity and since multilayer embedding is feasible, capacity can be more than 1 bit per pixel. Histogram modification is the second reversible data hiding algorithm introduced, leading to good image quality. Finally, data is embedded in high dynamic range astronomical images, for use of metadata binding. The embedding algorithm uses the Walsh-Hadamard transform and proved to be robust to FITS-to-TIFF lossless format transform.