基于线性预测倒谱滤波的音频水印检测技术研究

发布时间：2018-07-17 01:48

【摘要】：随着网络与多媒体技术的迅速发展,多媒体产品面临被轻易地复制、传输与修改的难题。加强媒体信息安全与保护媒体知识产权的需求日益迫切。数字水印是维护多媒体产品安全的一种有效解决方案,通过在媒体数据中添加代表版权所有权的隐秘信息,可以达到保护版权、鉴别数据真伪及标识产品等目的。在感知灵敏区域内,相比于视觉系统,人的听觉系统具有更高的灵敏度,因此,在数字音频信号中添加水印的技术比静态图像水印技术的难度更大。传统的扩频音频水印技术中,宿主信号在水印检测过程中充当着噪声源的角色,使得水印检测的误码率比较高。本文的研究内容就是为了降低水印检测误码率,同时保证音频的感知透明性,降低宿主信号对水印的影响。在音频水印技术领域,普遍采用减少信号方差来提高水印检测率。传统方法分别是Savitzky-Golay滤波、谱包络滤波、倒谱滤波和线性预测。尽管这四种方法都可以减少信号方差,但目前尚缺乏公开的文献资料分析这四种技术,缺乏实验代码实现和实验分析,因此对各项参数对实验性能的影响缺乏定量的数据分析和指导。本文通过Matlab实现了这四种降低信号方差方法的实验研究以及性能分析,并在此基础上提出了线性预测倒谱滤波方法,能够显著降低样本点幅度,进而降低信号方差。然后,本文实现了基于扩频技术的音频水印算法的优化,在扩频水印嵌入模块增加了音频信号的频域特性分析,用来平衡水印的嵌入强度,以此保证嵌入水印后的音频信号具有良好的感知透明性,并将降低信号方差的方法应用到音频水印检测系统中,降低宿主信号对水印检测的影响。最后,将提出的扩频水印优化算法(OSS)与Malvar和Florencio提出的改进扩频水印方法(ISS)进行了感知透明性与误码率的比较,并利用Matlab实验,验证了优化算法可以有效地降低信号对嵌入水印的影响,提高水印检测性能,相对于ISS,优化算法可以降低2%的误码率。
[Abstract]:With the rapid development of network and multimedia technology, multimedia products are faced with the problem of easy replication, transmission and modification. The need to strengthen media information security and protect media intellectual property is increasingly urgent. Digital watermarking is an effective solution to maintain the security of multimedia products. By adding secret information representing copyright ownership in media data, we can protect copyright, identify the authenticity of data and identify products. In perceptual sensitive region, human auditory system has higher sensitivity than visual system, so it is more difficult to add watermark to digital audio signal than to static image watermarking. In the traditional spread spectrum audio watermarking technology, the host signal acts as the noise source in the watermark detection process, which makes the bit error rate of watermark detection higher. The purpose of this paper is to reduce the bit error rate of watermark detection, at the same time to ensure the perceptual transparency of audio and reduce the influence of host signal on watermark. In the field of audio watermarking, reducing signal variance is widely used to improve watermark detection rate. The traditional methods are Savitzky-Golay filtering, spectral envelope filtering, cepstrum filtering and linear prediction. Although all four methods can reduce the variance of signals, there is still a lack of open literature and analysis of these four techniques, lack of experimental code implementation and experimental analysis. Therefore, there is a lack of quantitative data analysis and guidance on the influence of various parameters on the experimental performance. In this paper, the experimental research and performance analysis of these four methods for reducing signal variance are realized by Matlab. Based on this, a linear predictive cepstrum filtering method is proposed, which can significantly reduce the sample amplitude and then reduce the signal variance. Then, the audio watermarking algorithm based on spread spectrum technology is optimized, and the frequency domain characteristic analysis of audio signal is added to the spread spectrum watermark embedding module to balance the embedding intensity of the watermark. The method of reducing signal variance is applied to audio watermark detection system to reduce the influence of host signal on watermark detection. Finally, the proposed spread spectrum watermarking optimization algorithm (OSS) is compared with the improved spread spectrum watermarking algorithm (ISS) proposed by Malvar and Florencio. It is verified that the optimized algorithm can effectively reduce the influence of the signal on the embedded watermark and improve the watermark detection performance. Compared with ISS, the optimized algorithm can reduce the bit error rate by 2%.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309.7

【参考文献】