A Novel Audio Watermark Embedding and Extraction Method Based on Compressive Sensing, Sinusoidal Coding, Reduced SVD, Over Complete Dictionary and L1 Optimization

Abstract

Digital audio watermarking is relatively a new technology to stop audio piracy and to ensure security of the ownership rights of the digital audio data. In this paper, a novel digital watermark embedding and extraction method for audio data is proposed, satisfying the demands of robustness and imperceptibility. This method is based on sinusoidal coding of speech, Compressive Sensing (CS), Reduced Singular Value Decomposition (RSVD), Over-Complete Dictionary (OCD) matrix and L1 optimization algorithm. The sinusoidal approximation of original watermark signal is embedded into the compressive measurements of the host audio signal by using RSVD. Random sampling through compressive sensing ensures compression as well as encryption of the host audio signal. The extraction procedure is based on over-complete dictionary matrix and L1 norm optimization. The over-complete dictionary is created by using sinusoidal speech coding bases and compressive sensing measurement matrix. Experimental results show that proposed method provide exact recovery of watermark information and host signal under noisy attacks. © Springer-Verlag Berlin Heidelberg 2014.