Audio Perceptual Hashing Based on NMF and MDCT Coefficients

Audio perceptual hashing is a digest of audio contents, which is independent of content preserving manipulations, such as MP3 compression, amplitude scaling, noise addition, etc. It provides a fast and reliable tool for identification, retrieval, and authentication of audio signals. A new audio hashing scheme based on non-Negative matrix factorization (NMF) of Modified discrete cosine transform (MDCT) coefficients is proposed. MDCT coefficients, which have been widely used in audio coding,. exhibit good discrimination for different audio contents and highly robustness against content preserving manipulations, especially MDCT based compression such as MP3, AAC, etc. Based on the extraction of MDCT coefficients of the audio frames firstly, NMF is used to construct hash bits. Experiment results demonstrate that, compared with methods mentioned in literature, the proposed scheme exhibits a high efficiency in terms of discrimination, perceptual robustness identification rate and time consumption.