Research On Semi-fragile Watermarking Algorithms For The Verification Of The Integrity Of Digital Products

In recent years, digital multi-media has become increasingly popular and its influence has extended to every aspect of people's lives. Howerve, the advances in software technology have made it possible for almost anybody to maliciously or non-maliciously alter digital products, such as musics, videos and images. This has put digital products in an unprecedented credibility crisis. Against such a backdrop, watermarking techniques that can detect malicious tampers, if any, on digital products have become more and more attractive in a practical point of view to researchers and scholars from around the world.After briefly exploring existing watermarking theories and algorithms, a semi-fragile watermarking technique for image authentication was presented in this paper. By applying a pseudo-random permutation process to wavelet coefficients of the original image, tamper localization capabilities of the proposed scheme has been substantially improved. In the proposed scheme, the watermark bit was used to monitor a random area of the original image. By so doing, tampering localization capabilities of the technique has been increased while the watermark payload has declined. Furthermore, since the watermark bit was embedded in the wavelet domain, the watermarked image is intrinsicly secure to block-based local attacks, such as vector quantization and cropping. Theoretical analysis and experimental results have demostrated the performance and effectiveness of the proposed algorithm.A semi-fragile watermarking algorithm based on image hashing is also proposed in the thesis. A novel way to compute image hash value is firstly presented after exploring the difference between cryptographic hashing and image hashing. In order to embed the watermark signal, the original image was first divided into sub-images of predetermined size. Then the hash value of every sub-image was embeded in two other sub-images, the locations of which were controlled by a private key. In this way, a doubly linked chain was formed. By computing the hamming distances between the original hash value and the two extracted hash values, the system can successfully detect and locate areas that were maliciously attacked.The third image authentication technique presented in this document was based on Human Visual System(HVS) and Discrete Wavelet Transform(DWT). The proposed algorithm has taken full advantage of the HVS characteristics for the incorporation of HVS masking image in the watermark embedding process. Arnold transform was first applied to the binary watermark image to get a chaotic image of the watermark image. Then the wavelet coefficients of the original image were updated according to the watermark sequence and the wavelet coefficients of the HVS masking image. The proposed algorithm can automatically classify the type of changes on the watermarked image. Simulated experiments have demostrated that the proposed system is robust to content-preserving manipulations, but highly fragile to malicious changes on the watermarked image.