Research On Blind Image Watermarking Algorithm Based On Visual Model

With the rapid development of the information technique and computer network, the storage, copy and distribution of the digital multimedia, such as image, text, audio, video and three-dimensional model, is becoming more and more convenient, and people's life and work is becoming to be digitalized and network-connected. We can not only obtain the duplicates from the internet, and also can edit, copy and distribute these digital products. Information security, pirate and copyright problem have come to be society problem. For this background, the digital watermarking was developed, which protect the digital product by embedding some secret information into the production. Nowadays, the digital watermarking still has some problems to be solved, such as robustness and fidelity. We know that the two aspects conflict with each other. However the embedding based on the visual model could achieves tradeoff between them. The works in this thesis are as follow:1. A new robust image watermarking scheme based on Independent Component Analysis (ICA) is proposed. In the proposed scheme, the original image is sub-sampled firstly to obtain four sub-images, and the features of the original image are extracted by performing ICA to these sub-images, and then the feature, which has the largest variance, is selected out to be watermarked on its block DCT coefficients. During the embedding, Human Visual System (HVS) is used to determine which regions should be watermarked in order to balance the imperceptibility and the robustness. Experiments show that the watermark can be extracted blindly, and it is robust to common watermark attack and image processing, such as JPEG compression, noising, scaling, filtering, histogram equalization, and so on.2. A nonlinear quantization scheme is proposed to conquer the drawbacks of the uniform quantization step of the conditional QIM. The scheme selects a nonlinear function according to the Human Visual System. First, the value to be quantized is transformed by this nonlinear function. Second, the watermark is embedded into the value by traditional QIM with the uniform step. Finally, the watermarked value is obtained by inverse function of the nonlinear function. Using this scheme, different value can be quantized by different steps. The proposed nonlinear Quantization Index Modulation (QIM) algorithm based on visual model ensures the watermark's imperceptibility as well as its robustness.3. Adaptive Quantization Index Modulation (AQIM) proposed by Cox calculates the quantization step according to the Watson's visual model, which makes the quantization step adaptively vary with the coefficient to be quantized. It has better reliability and robustness than the traditional Quantization Index Modulation. But in the AQIM algorithm, the embedding step during watermarking is different from the one in the watermark detection because of the blind detection. And this will cause the trouble that the watermark couldn't be extracted integrally. In this paper, an iterative AQIM scheme is proposed based on the analysis of this algorithm, which has better performance than the original AQIM.