摘要
本文分析测试了影视音乐音频盲取技术的环境监测,在提取过程中所使用的数学工具包括小波包分析、梅尔倒谱系数等,综合探究其特征性,同时结合期望最大化算法展开机器训练聚类。结果表明,影视音乐音频录制环境的分类准确率获得大幅提高;小波和傅里叶的特征提取,最终所得到的k均值分类准确率低于期望最大化算法。将期望最大化分类算法与k均值相结合,与傅里叶算法的小波包特征后分类结果相比,前者的显然更优。
In this paper,we analyze and test the environment detection in audio blind forensics of film and television music,extract the feature by Mel cepstrum coefficient and wavelet packet analysis,and combine it with expectation maximization algorithm for machine training clustering.The results show that the classification accuracy of audio recording environment is greatly improved,and the k-means classification accuracy is lower than the expected maximization algorithm in wavelet and Fourier feature extraction.Using k-means and expectation maximization classification algorithm,the classification result after wavelet packet feature is better than that of Fourier algorithm.
引文
[1]程哲.音乐、音响在影视后期制作中的运用及作用[J].现代职业教育,2017(24):145.
[2]徐俊瑜.数字视频被动取证技术研究[D].天津:天津大学,2013.
[3]KIM J,HAN M M.A nated host detection and identification algorithm based on port patterns of syn packets[J].International Information Institute(Tokyo).Information,2014,17(1):271.
[4]李鹏超,杨鹏.基于Android手机的音频文件取证技术研究[J].网络安全技术与应用,2017(11):147-148.
[5]孙科建.基于模式噪声一致性的视频取证技术研究[D].徐州:中国矿业大学,2014.