Research On Audio Watermark Detection Technology Based On Linear Predictive Cepstrum Filtering

With the rapid development of network and multimedia technology, multimedia products are faced with the problems that are easily copied, transmitted and modified.There is an increasing need to strengthen the security of media information and to protect media intellectual property. Digital watermarking is an effective solution for maintaining multimedia product security,by adding the secret information that represents the copyright owner in the media data, it is possible to achieve the purpose of protecting the copyright, authenticating the authenticity of the data, and identifying the product.In the sensitive area, compared with the visual system, the human auditory system has a higher sensitivity, therefore, the technology of digital watermark embedded in audio signals is more difficult than static image watermarking. In the field of traditional spread spectrum audio watermarking technology, the host signal acts as a noise source during the watermark detection process, resulting in a higher watermark extraction error rate. Therefore, the research content of this paper is aimed to reduce the bit error rate of watermark detection, while ensuring the perception of transparency audio, reduce the impact of the host signal on the watermark.There are four traditional methods to improve the watermark detection rate by reducing signal variance. They are Savitzky-Golay filter, spectral envelope filtering,cepstrum filtering and linear prediction method, respectively. Although these four methods reduce the signal variance, there is still a lack of open literature to analyze these four technologies. The lack of experimental code and data analysis hinders the acting of selecting parameters for audio watermarking. In the thesis, we achieved the experimental study and performance analysis for the four signal variance reduced methods with Matlab experiments. A linear predictive cepstrum filtering method is proposed based on our analysis results, which can significantly reduce the sample point amplitude and the signal variance.Then, the thesis proposes an optimal audio watermarking algorithm based on spread spectrum technology. The frequency domain characteristic of the audio signal is introduced to spread spectrum watermark embedding module to balance the embedding intensity of the watermark, so as to ensure good perceived transparency after watermark embedded. Furthermore, we apply this method of reducing signal variance into the audio watermark detection system.Finally, the proposed optimal method is compared with the improved spectral watermarking algorithm (OSS) proposed by Malvar and Florencio in terms of perceived transparency and robustness. Experiment results verify that our algorithm can effectively reduce the influence of signal on the embedded watermark and improve the performance of watermark detection. Compared with ISS,the proposed algorithm can reduce the bit error rate by 2%.