摘要： 基于二进制哈希的音频指纹匹配方法鲁棒性较差。为此，提出一种帕尔森高斯核量化哈希方法，将2个音频中间哈希值的差异度通过概率密度函数量化编码到一个合适的整数范围内，以刻画失真的概率分布，实现音频指纹的提取。实验结果表明，与二值哈希法相比，该方法对多种失真具有更高的鲁棒性。

关键词: 音频指纹, 指纹提取, 量化哈希, 帕尔森高斯核, HFM方法

Abstract: Aiming at the poor robustness problem of the binary hash for the audio fingerprint matching technology, this paper proposes a Parzen Gaussian kernel quantum hash scheme, which encodes the intermediate hash difference between two audio contents into an integer at a suitable range according to its probability density function, and characters the probability distribution of distribution. Experimental results show that the method is more robust under various distortions than the binary hash.

Key words: audio fingerprint, fingerprint extraction, quantum hash, Parzen Gaussian kernel, HFM method

中图分类号: 

• TP391