| dc.description.abstract | The advances of digital techniques and Internet have promoted the style of contents from physical form to digital form. Digitized contents can be distributed and duplicated rapidly without any degradation. This property is both an advantage and a drawback of digital technique. Pirate may distribute or modify digitized contents without the permission of the creator or owner. This behaviour will be of no doubt a threat to daily digital life. The objective of this dissertation is to investigate how to protect the ownership and contents of digital images using the technique of digital watermarking. According to different purposes, watermark technology can be categorized into robust, fragile, and captioning watermarking. In this dissertation, we will propose techniques to solve the problem of copyright protection, content authentication, and copied ID card authentication using robust, fragile and captioning watermarking, respectively.
In the first part of this dissertation, we will design a message-based robust watermarking system which is much more flexible than conventional noise-type and logo-type watermarking system. Usually, user can input a message-type watermark directly through a keyboard and this flexibility makes a message-based watermarking system a great copyright protection tool. The proposed system consists of a robust cocktail watermarking modulation scheme and a powerful coding mechanism which integrates Hadamard code and backpropagation neural network. Experimental results reveal that the embedded message can be reliably extracted even if the cover image is attacked by image processing.
The second part of the dissertation will focus on developing a powerful image authentication system which can detect malicious tampering while tolerating incidental modifications. Since a fragile watermark can be destroyed easily by small amount of modification, we can use this property to detect the area in which a malicious attack has been imposed. However, fragile watermark can not tolerate incidental distortion also. To detect the area which are maliciously tampered with and to tolerate the modification which is applied incidentally simultaneously, the semi-fragile watermarks can be applied to achieve the goal. For achieving this goal, we propose a semi-fragile watermarking system based on a mean-quantization based watermarking approach which increase the robustness of conventional quantization-based fragile watermarking. We derived a method to determine the most suitable amount of robustness and fragility of semi-fragile watermark such that the embedded watermark is robust to incidental distortion and fragile enough to detect the area which is tampered with. Experimental results reveal that the proposed method is effective in distinguishing an instance of malicious tampering from a incidental distortion.
In the third part of this dissertation, we try to investigate the practical applications of a watermarking system. We propose a copied ID card authentication system based on face watermarking. The proposed system authenticates an copied ID card by checking the integrity between the clear text shown in ID card and the hidden text embedded in a photo image on ID card. The proposed system adopts a morphology-based face detection technique to detect the face on an ID card. Then, the detected face image is then used as side information during the watermark embedding/extraction for achieving blind detection purpose. The copied ID card is regarded as authentic if the hidden text is consistent with the explicit one.
Otherwise, the copied ID card is treated as a fake one. Preliminary experimental results are provided to illustrate the concepts. A complete system will be implemented in the future work. | en_US |