Summary: | 碩士 === 國立清華大學 === 資訊工程學系 === 87 === A novel image watermarking method with desired picture quality is proposed in this thesis. As usual, the watermark information is a long sequence of random values of normal distribution with zero mean and unity variance. The embedding process operates on the DCT domain as well as the spatial domain regarding the promotion of robustness and invisibility together. First, we understand that the quantity of embedding information is limited by the quantity of the corresponding DCT coefficient, especially for the low quantity coefficient the capability of robustness is declining. Conversely, it can sustain large quantity of information with pleasing level of robustness no matter how the postprocessing is done. Accordingly, our watermarking method adopts multiple and adaptive scaling factors to effectively embed the watermarking information with larger capacity relative to the other methods can do. These scaling factors are guided by desired image quality also. Besides, we superimpose a JND visual model on our method in spatial domain to guarantee the invisibility of watermarks. Furthermore, after the destructive postprocessing such as cropping or resizing, the current similarity measurement is useless. We propose an interesting similarity measurement that can truthfully reflect the similarity using the partial set of watermarks we only have. Finally, we conducted several experiments to justify the proposed method. The simulation results show that it surely prevent the original quality and against the JPEG compressing, cropping and resizing.
|