| With the constant upgrading of electronic equipment and social platforms in the information age,multimedia file as the carrier of information plays an important role in it.Security as a hot topic of media attention,including tracking authentication,copyright protection and other issues have become the driving factors in the field of information security protection.Digital watermarking has been widely studied as an effective means of information because of its effective copyright protection,platform compatibility and wide application of traceability.Digital watermarking technology is a kind of information hiding technology in which the watermark information is embedded into the appropriate position of image pixel value or frequency domain coefficient.In recent years,due to the rapid development of Just Noticeable Difference(JND)and the high robustness of watermarking based on Discrete Wavelet Transform(DWT),robust watermarking based on JND model has been widely used in image copyright protection.Generally speaking,in the existing methods based on JND model,the characteristics based on transformation domain play an important role in the quantization strength measurement process.However,most existing methods only use the subband-specific features in the final multiscale wavelet transform domain,and ignore the feature correlation between subband-specific features at different scales.In order to solve this problem,we have proposed a JND estimation mechanism based on cross-scale feature fusion for the establishment of robust image watermarking.In addition,the existing saliency model based on traditional spatial domain calculation has low robustness in the process of watermark embedding and extraction by taking advantage of the differences in the allocation of human brain’s attention to different regions of the image.It becomes particularly important to realize the combination of effective significance estimation and visual masking in the wavelet domain,so we also propose the JND estimation mechanism of significance modulation for the establishment of robust image watermarking.The main work is as follows:1.In this thesis,for the proposed problem of visual masking estimation for embedded specific subbands,different quaternion wavelet scales are first designed for JND modeling,which can be used to measure the correlation between the minimum distortion levels of embedded specific subbands.In addition,cross-scale features between different subbands are used to jointly generate texture and pattern-guided contrast masking measurements.Then in this thesis,channel coefficient components in the quaternion wavelet domain are used to effectively extract cross-channel color information for the construction of color complexity masking,making the final perceptual masking more representative.2.By studying the visual features of different channels in the quaternion wavelet domain,visual sensitivity is represented by the difference of the attention distribution of human eyes to different areas of the image.Based on the global likelihood of the feature and the difference of the central local area,a significance model with the integration of global and local salient features is established.Then,the value of quaternion wavelet domain significance corresponding to the image is used in the modulation calculation of the the embedding-specific subband visual masking,which is conducive to obtain the best quantization intensity of the embedding-specific subband.3.We further develop a robust image watermark based on the cross-scale fusion JND estimation mechanism and significance modulation in the quaternion wavelet domain,which embed the watermark by quantizing the candidate quaternion DWT coefficients.The coefficient is defined adaptively according to the proposed visual masking value.Experimental results based on open image databases show that compared with other state-of-the-art image watermarking methods,the proposed method can produce reliable and promising results. |
PDF Full Text Request