Research On Publicly Sensitive Watermarking Based On Multi-level Wavelet Coefficient Weighting And Equal-band Fusion

Digital watermarking technology as an important method of copyright protection and integrity authentication has become a hot research topic in the field of information security. Conventional watermarking technique usually paid attention to the robustness and invisibility of watermarking algorithm while ignored the security of watermarking. Some spatial domain watermarking algorithms used key to disturb the embedding position of watermarking or watermarking scrambling, encryption, but they usually embed the watermark in not significant bits and was easily to be eliminated. Although some frequency domain watermarking algorithms protected watermarking to a certain degree, but using fixed frequency domain transformation led to fixed watermarking embedding positions which are prone to be attacked. The experiment confirmed that the sensitivity of watermarking algorithms based on parameterized wavelet transform were difficult to be applied in practice. Some non-blind watermarking algorithms embedded the watermarking into visual important low frequency coefficient to improve the robustness and security of the watermarking embedding stage, but it needs the original carrier when extracting watermarking. In addition, the existing watermarking algorithms usually have nothing to do with the attributes of cover or watermarking image, different covers or watermarking images experienced the same embedding and extraction stage, which led to the poor security of watermark embedding.To address these problems, the followings are finished works:1) A public sensitive watermarking algorithm combined weighted multilevel wavelet coefficient with mean and quantization was proposed. In the proposed algorithm, the MD5 value of cover image, user keys and initial parameters were bound with chaotic map which were used to produce binary pseudo random matrix for binary watermarking xor encryption, and generate the middle parameters which were used to select wavelet coefficients randomly in different decomposition levels in the embedding stage. Then the weighted multilevel wavelet coefficient mean was adjusted to embed watermarking. In the extraction stage, an isolated point filtering strategy was used to enhance the quality of fetched watermarking. Compared with the traditional watermarking algorithm, the proposed algorithm was blind watermarking embedding algorithm, it doesn't need to provide the original image in the embedding stage, and different cover image MD5 value, user keys or initial parameters corresponding to different watermarking encryption, embedding and extraction stage. The experiments show that the proposed method has better performance in plaintext image and user keys sensitivities. The embedded watermarking is difficult to be extracted or tampered even if all strategies of the algorithm are published. And it is still robust for image attacks such as random noise, JPEG compression, image clipping, covering and graffiti.2) The proposed public sensitive watermarking algorithm combined weighted multilevel wavelet coefficient with mean and quantization still exist following problems: ? The performance difference of wavelet coefficients at all levels led to quality decrease of the extracted watermarking; ? An isolated point filtering strategy was used to improve the ability to resist attacks of watermarking;? Each stage of the watermarking algorithm was only related to the attributes of cover image, and had nothing to do with attributes of watermarking, which led to the risk that the same cover embedding different watermarking images experienced the same watermarking encryption and embedding process. Therefore, a public sensitive watermarking algorithm with non-overlapped block and wavelet coefficient fusion of equal frequency band is proposed. In the proposed algorithm, firstly the SHA-1 value of cover image and watermarking image, user keys were used to generate initial values and parameters of watermarking encryption stage and the starting point coordinates of random non-overlapping blocking of watermarking embedding stage. Then l level wavelet pyramid decomposition on each block was done, and HL_l, and LH_l, frequency band coefficients were fused with the same weight to form fusion coefficient. After that, the fusion coefficient was adjusted to embed watermarking. The proposed algorithm avoided the short board effect which was brought by the fused wavelet coefficients at different levels, and it also avoided the isolated point filtering strategy for extracted watermarking. Each stage of the watermarking algorithm was not only related to the attributes of cover image, but also was closely related to the attributes of watermarking. It provided higher safety and better practical application value. The experiments show that the proposed method still provides enough security and the original watermarking can be used to authenticate watermarked image. It is also robust for image attacks such as random noise, JPEG compression, image clipping, covering and graffiti.