Research On Perception Invariant Feature Oriented Media Authentication Techniques

Images (static and dynamic) are the most important multimedia for information propagation and knowledge accumulation. The wide applications of the huge amount of images have brought new challenges to multimedia content authentication. Water-marking and robust Hash are two promising techniques for multimedia security. How-ever, those two technologies do not have strong theoretical basis and are not well studied in application domain. As such, existing algorithms cannot satisfy the specific require-ments of applications. Therefore, we are motivated to design new algorithms according to specific applications, which would greatly improve the recent multimedia authentica-tion techniques. To solve the existing problems in previous research work, we analyze in details the requirements, developments, and theoretical basis of recent watermark and robust hash techniques. Based on the visual perception theory, we have greatly improved the existing watermarking and robust hash algorithms in performance and security. In this thesis, we have made the following contributions:(1) In requirements analysis, we propose to balance multiple targets rather than emphasizing single performance. We adopt human visual perception theory in algorithm design to simulate perceptual information processing. As there has been no standard evaluation index for Robust Hashing algorithms for long, we analyze in half-quantity and quality the robustness, discrimination, one-way property, and security of Robust Hashing algorithms. Moreover, we compare the paradigm of visual perception and multimedia security techniques, based on which we propose the concept of perceptual hash based on property combination and soft match。And finally, we discuss the future direction of perceptual hashing.(2) We have proposed a watermark algorithm based on the invariability of SVD parameter correlation. Our algorithm overcomes the deficiency of previous algorithms which cannot distract image content blindly. Moreover, our algorithm is self-adaptive to embed different multimedia contents, which is robustness for attacks with high image quality.(3) We have proposed another watermark algorithm based on FrFT according to time-frequency distribution theory. This algorithm is self-adaptive and robustness for compression attacks.(4) We have proposed a robust hashing algorithm based on image luminance dis-tribution. The luminance information is extracted by Radon transformation, and the dis-tribution correlation of different luminance sections are described by wavelet transform. The hash value is generated by Fourier transform. As the algorithm is based on the corre-lation of perceptual properties, and adopts the invariability of Radon transform, multiple discriminations of wavelet transform, one-way property of Fourier transform, it has great robustness, discrimination, and security properties.(5) We have proposed a perceptual hash algorithm of combined properties based on multi-resolution processing strategy of visual perception information. This algorithm combines the advantages from the invariability of luminance information and accuracy of structure information, and makes up for the attacked content through template match. As such, we have greatly improved the algorithm quality. And we propose another hash algorithm based on the feature extraction strategy. This algorithm combines detail fea-tures and geometric features, which has great robustness, discrimination, as well as key-dependency and tamper detection.