Study Of Digital Image Watermarking In Frequency Domain And Image Quality Evaluation

With the popularity of network communications and digital media information, the transmission of digital multimedia information becomes much more convenient and quickly. But the security issue of digital media content has become a choke point, which constrains its applications. As a potential means to solve this problem, digital watermarking becomes a common focus of industry and academia society in recent years.Firstly, the principle and the framework of digital watermarking are analyzed, and some key techniques are introduced also. The basic principle of watermarking in DCT domain is introduced. A blind digital image watermarking algorithm in DCT domain is proposed. No original image is needed when detecting the watermark. The embedding position of the watermark is determined by a key and the security of the system is enhanced.Secondly, the theory of wavelet transform, as well as its application in image processing, is analyzed in detail. The idea of self-adaptive watermark embedding is introduced. An adaptive gray-level digital watermarking algorithm is then proposed. Several masking properties of the HVS (Human Visual System) are combined with the watermark embedding, such as the illumination of masking properties and texture masking properties. The algorithm consists of the following three steps: preprocessing of watermark sequence, embedding and extraction. Two important steps of the preprocessing are scramble and encryption. The embedding procedure will be established according to HVS of visual masking model, which adaptively embed watermark sequence. The extraction is the inverse of the embedding. The algorithm is robust to some attacks while keeping the watermark imperceptible. The attacks include low-pass filter, median filter, etc. The experimental results show: the robust of watermarking system is more closely related to the illumination of the mask properties, and texture features properties can improve the performance of the noise resistance for the watermarking system.Finally, a new objective image quality evaluation method, WEWMSE (Wavelet and Energy-Weighted Mean Square Error), is presented based on DWT. The sensitivity of HVS is weighted using the energy of sub-band. In this way, the assessment values of the new method agree well with MOS (Mean Opinion Score). In order to demonstrate the effectiveness and to verify the reliability of WEWMSE method, a series of experiment are made on an images library of JPEG and JPEG2000 with MOS labeled. The results show that the method more reliable, effective, and closer to the subjective evaluation results than the traditional method of objective evaluation,such as MSE and PSNR.