Digital Watermarking Robust Against Synchronization Attacks

One of the key points that to be more improved is Robustness against loss of synchronization which restricting the digital watermarking used in practice. It has received a lot of attention in recent years.The innovative pursuits in the dissertation can be summarized as the following four aspects.Firstly, the concept of self- synchronization audio watermarking based on reverberation is proposed. The watermark is embedded in the process of adding reverberation to recordings in music studio. The reverberation that applied to modulate the watermark generated by taking a portion of the original signal within a specified time and introducing a slight modification by amplitude shifting according to a secret key. If time and pith scaling manipulations are performed, these transformations occur in the watermarked and the replica generated from this watermarked signal simultaneously. The correlation is invariant against this type of manipulation applied to both signals, thus the scheme is robustness against synchronization attacks. It is a self-synchronization one with immunity against synchronization attacks. The imperceptibility is achieved by choosing the matched parameters with the audio style to get the best effect of art. The parameters for calculating the room impulse response are used as secret key that must be used in the decoding process to ensure security.Secondly, a kind of weight Hausdorff distance is defined. It is applied to measure the similarity of original and distorted image in order to achieved distorted watermarked image correction. It is usually that the distorted watermarked image is different from original one such as transformation, rotation, scaling. The extracted feature points are uniform absolutely is impossible. Hausdorff distance is a kind of distance between two point sets. It can ignore the inaccurate of the location of feature points. Thus, the correction scheme based on Hausdorff distance is fit for distorted watermarked correction. The experiment shows that the correction precision is better than other correction method.Thirdly, prediction error filter used in spatial watermarking to construct perceptual mask and pretreatment before normalized correlation detection. The perceptual mask that constructed by prediction error filter is a good descriptor of local image properties and well suited to the HVS. It applied to control the intensity of embedding. The experiment demonstrates that the proposed masking offers higher objective quality metric than HVS and NVF. Prediction error filter also applied to pretreatment normalized correlation detection in order to estimate the masked watermark. The experiment demonstrates that the performance was improved more than traditional correlation detection and pretreatment by whiten filter.Finally, a kind of naturally robust watermark template is applied to resisting the local and small nonlinear distortion. The big distortion is corrected by image registration based on Hausdorff distance. Combined with the prediction error filter, an algorithm robust against geometric attacks is proposed. The algorithm also robust against print-scan attack, it can be used preventing certificate forgery. This dissertation consists of seven chapters.In chapter one, describe the watermarking robustness, security and attacks, explain the synchronization attacks in detail. A kind of audio and image synchronization distortion model is given. Summarize detailed the audio and image algorithms against synchronization attacks.In chapter two, described the evaluating issues of audio and image watermarking. We require some evaluating and comparison method to study watermarking technique. A few evaluation criterions are used widely for audio and image are introduced in this chapter. In chapter three, study a kind of self-synchronization audio watermarking algorithm on the based on reverberation. The ideal of the scheme and the reverberation applied to modulate the watermark signals are introduced firstly. After describing the embedding and extracting process in detail, the experiment results are given.In chapter fourth, study the distorted watermarked image geometric correction schemes. The definition of basic Hausdorff distance and it's modified form are introduced firstly. A kind of weight Hausdorff distance is defined and applied to measure the similarity of original and distorted watermarked image to correct the distorted watermarked image. Then, three methods that extract the image feature points are introduced. The distortion model is given. A fast divide and conquer strategy in six dimension space is described. Two experiment results show that the scheme is a precision one.In chapter fifth, a kind of robust against geometric watermarking scheme in special domain is proposed. Natural robust watermark pattern is used to counter the small local linear and non-linear geometric distortion. The big distortion is corrected by image registration based on Hausdorff distance. Prediction error filter applied to construct masking function and pretreatment in normalized correlation detection. Four experiments demonstrate respectively the masking quality of the perceptual mask constructed by prediction error filter, the validity of the natural robust watermark, the performance of the pretreatment of the prediction error filter before detection and the robustness of the watermarking algorithm.In chapter sixth, a watermarking scheme robust against print-scan is proposed, which could be used in preventing certificate forgery. The attacks introduced during print and scan is analyzed firstly. Summarize the algorithms against print-scan attacks. Then a kind of scheme that aims at preventing certificate forgery is given. The two experiments used standard image and head picture sheet as carrier image respectively. The experiments results are analyzed in detail to get the conclusion that the scheme can be used in preventing certificate forgery. In chapter seventh, a brief summary of the dissertation is given. The suggestion for future researches is put forward.