From blind to non-blind audio watermarking: A new approach based on blind source separation

Abstract

This paper presents a new blind multi-watermarking scheme based on BSS (Blind Source Separation). At the emitter, multiple watermarks are embedded using CDMA (Code Division Multiple Access). At the receiver, we consider the detection problem of the multiple watermarks as a matter of BSS between the audio signal and the watermark one. We make our choice on the UISA (Under-determined Independent Subspace Analysis) technique that has proven its performance in terms of detection reliability. Recovering the estimated audio signal at the receiver, we show how to transform the blind detection to a non-blind one. Experimental results show how the detection of multiple watermarks can be improved thanks to recovering the estimated signal at the reception.