Using Low-Order Auditory Zernike Moments for Robust Music Identification in the Compressed Domain

Methods based on moments and moment invariants have been extensively used in image analysis tasks but rarely in audio applications. However, while images are typically two-dimensional (2D) and audio signals are one-dimensional (1D), many studies have showed that image analysis techniques can be successfully applied on audio after 1D audio signal is converted into a 2D time-frequency auditory image. Motivated by these observations, in this chapter we propose using moments to solve an important problem of audio analysis, i.e., music identification. Especially, we focus on music identification in the compressed domain since nowadays compressed-format audio has grown into the dominant way of storing and transmitting music. There have been different types of moments defined in the literature, among which we choose to use Zernike moments to derive audio feature for music identification. Zernike moments are stable under many image transformations, which endows our music identification system with strong robustness against various audio distortions. Experiments carried out on a database of 21,185 MP3 songs show that even when the music queries are seriously distorted, our system can still achieve an average top-5 hit rate of up to 90% or above.