多格式音频感知哈希算法

提出一种基于双树复小波变换的多格式音频感知哈希算法，解决了现有音频认证算法音频格式单一、算法不通用、效率低的问题. 首先对预处理后的音频信号进行全局双树复小波变换，获得信号的实小波和复小波系数，对它们分别分帧，帧数相同；对实小波系数计算每帧信号Teager能量算子的模值，作为实小波系数的帧间特征，接着对每帧信号再分帧，提取再分帧帧信号的短时能量作为实小波系数的帧内特征；对复小波系数求取每帧信号的熵值作为复小波系数的帧间特征；最后对上述特征分别进行哈希构造，生成感知哈希序列. 实验结果表明，该算法对5种不同格式的音频都具有强鲁棒性，且区分性好，效率高，并能实现小范围篡改检测.

Perceptual Hashing Algorithm for Multi-Format Audio

A novel multi-format audio perceptual hashing algorithm based on dual tree complex wavelet transform ( DT-CWT ) was proposed. It solves the problems of the existing audio authentication algo-rithms, including that audio files are kept in a single format, and algorithms are not generic and low effi-ciency. The proposed algorithm first applies the global DT-CWT to the audio signal after pre-processing conducts to obtain the real and complex wavelet coefficients. Next, the coefficients are partitioned in some frames respectively, and the frame numbers are same. For the real wavelet coefficients, the module values of teager energy operator in every frame are computed to serve as its inter-frame feature. And then short-time energy of the new signal, which is generated to frame the frame signal, is computed to serve as its intra-frame feature. For the complex wavelet coefficients, entropy values are obtained in every frame to serve as its inter-frame feature. Finally, the above features are to conduct a hashing structure process to produce the perceptual hashing sequence. Experiments show that the proposed algorithm has good robust-ness and discrimination for audio signals with five different formats, with high efficiency and ability to im-plement tamper detection as well.