An intelligent audio watermarking based on KNN learning algorithm

The quick development and advance in information technology and computer networks have brought a more and more attention to data transmission in digital form. The main problem of owners and producers of digital products is to defend against distribution and unauthorized copying. An efficient solution for this disturbance is using the digital watermarking techniques. The purpose of audio digital watermarking is to insert a series of hidden information into audio files, so that it can’t be heard and is robust against signal processing attacks. One of the major problems of the conventional audio watermarking schemes includes the rule-based decoders which use some sets of specific rules for watermark extraction without any intelligence. In this paper, we propose a new robust intelligent audio watermarking scheme based on a collaboration of discrete wavelet transform and K-nearest neighbor (KNN) techniques. The scheme carries out the embedding of watermark data based on modifying energy levels in wavelet domain. Moreover, an intelligent KNN learning machine is trained to capture the correlation between modified frequency coefficients, in wavelet domain, and the watermark sequence. In extracting phase, the watermarked data can be effectively retrieved using the trained KNN machine. The watermarking scheme preserves data synchronization through inserting a chaotic sequence. In order to evaluate imperceptibility and robustness of the proposed watermarking scheme, several experiments under various conditions are carried out. The experimental results show a relatively better imperceptibility, higher robustness and capacity than the conventional techniques. Data embedding rate is 1600 bps.