Integration of Multiple-Description Iterative Coding and Inter-Block Spread Spectrum in Image Watermarking System

• Main Authors: Ying-Fen Hsia, 夏英峰
• Other Authors: Jan-Ray Liao
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/88743739227524731574

博士 === 國立中興大學 === 電機工程學系所 === 98 === To embed the multiple watermarks is a way to increase robustness in image watermark. The limitation of embedding multiple watermarks is that the limited capacity of an image severely limits the size of the watermark. Multiple-description coding is a good candidate to solve this limitation by trading off between transmission bandwidth and bit error rate. Traditionally, multiple-description coding is considered in on–off channels where channels are not marred by bit errors but occasional connection outages such as dropped packets. To apply multiple-description coding in image watermarking, we need a form of multiple-description coding for noisy channels instead of on–off channels. In this thesis, we propose to add iterative coding in multiple descriptions in order to combat bit errors in image watermarking. We call this method multiple-description iterative coding (MDIC) image watermarking. We tested our system on six images. On average, bit error did not happen until we compressed the image in JPEG to PSNR 36.97 dB. We concluded that MDIC was a very good way to increase robustness for image watermarking. One of the most popular watermark embedding methods is the direct sequence spread spectrum (DSSS) method. The original DSSS technique divides the image into NxN blocks. These blocks are transformed by discrete cosine transform (DCT). Then, the pseudo-noise (PN) sequences are embedded in selected DCT coefficients within the blocks. In this thesis we explore another two aspects in spread spectrum DCT-based watermarks that are yet unexplored in previous researches. The first aspect is what we called “inter-block” method. In previous DSSS watermark algorithms, the PN sequence from one input bit is embedded in one block. We called this approach “intra-block”. On the other hand, we can embed the PN sequence into several different blocks. Thus, we called it “inter-block” method. The rationale for this “inter-block” approach is that it spreads the same bit into a wider region in the image. This wider spreading may be beneficial in watermark counter-attack. The second aspect is the use of frequency-hop spread spectrum (FHSS). Previous researches indicated that FHSS performed better than DSSS in certain channel condition but FHSS has only been used in audio watermarks. Therefore, applying FHSS watermarks in images is worth investigating. To study the above two aspects, we developed three image watermark algorithms. They are called “intra-block FHSS”, “Inter-block DSSS (IDSSS)” and “Inter-block FHSS (IFHSS)”. These three methods and the original DSSS or “intra-block DSSS” were compared by subjecting them to various watermark attacks. From the experiments, we have the following findings. “Inter-block” techniques perform better against all attacks. For filtering and quantization attacks, IDSSS watermarks performed better. For geometric distortion attacks, IFHSS watermarks performed better. In order to utilize the advantages from these two different approaches, we combined IDSSS and IFHSS into a single watermark system, called Inter-Block Combined (IBC) system, and tested it. The results showed that the IBC can resist a broad range of watermark attacks. Finally, we combined the above two systems, MDIC and IBC, together to construct a whole new architecture, called combined MDIC or CMDIC watermarking system. The experimental results showed that CMDIC can defend a wider scope of watermarking attacks.