Detection And Active Attack Against Hidden Information In Digital Images And Audio Signals

With the development of communications, signal processing and computer networks, especially the seamless connection of the Internet, information hiding has attracted much attention in recent years. Digital watermarking for copyright protection has been studied intensely, and many products have been developed. In the mean time, hidden communication through subliminal channels is also being investigated, which transmits secret information within multimedia data in public networks. This emerging technological field is closely related to the important issue of information security.This thesis investigates information hiding techniques aimed at hidden communication. A general discussion on the background is first given. A literature survey on the related research and recent development is then presented. The very nature of modification to the host data due to information hiding is explored, leading to effective analysis of, and attacks against, some commonly used data embedding schemes. The main contributions and innovative results of this work include:1. As a basis for detection of subliminal channels, an important measure representing invisibility of the hidden information in still images is studied. After analyzing the drawbacks of the conventional mean square error (MSB) method, an improved metric, the weighted mean square error (WMSE) is proposed. Theoretical and experimental studies show that, when used to measure image distortion caused by data embedding and image processing, WMSE is closer to the human visual assessment than the original MSE. The WMSE metric is particularly useful in evaluation of data hiding algorithm in terms of invisibility.2. For blind detection of the existence of a watermark or secret communication, this thesis proposes three criteria based upon analysis of local statistic properties in the data for making decision as to which of the two given versions of the same multimedia material (digital audio or image) carries secretdata. These three criteria are mutually complementary. Experimental results are presented showing the effectiveness of the method.3. An effective attack scheme is proposed against a commonly adopted data embedding technique, quantization index modulation (QIM). The described scheme not only removes the embedded information, but also, to some extent, compensates for the distortion in the host media due to embedding. This technique can be used as an active warden to prevent hostile hidden communication.In the final part, the thesis briefly discusses the benefits of combining modern encryption with steganographic techniques in enhancing security of information systems, and gives a perspective view to the future development of the field.The work was supported by Natural Science Foundation of China, Key Disciplinary Development Program of Shanghai, and Doctoral Programs Foundation of Shanghai.