A Novel Semi-fragile Audio Watermarking Technique Based on Support Vector Regression

摘要

Semi-fragile watermarking methods aim at detecting unacceptable malicious manipulations, while allowing acceptable regular manipulations. In this paper, a new semi-fragile audio watermarking algorithm based on the support vector regression (SVR) is proposed. This algorithm extracts the more steady features and adopts a new embedded strategy to resist the synchronization attack effectively. Firstly, the corresponding feature are selected as train sample, which are extracted from the audio segments after Discrete Wavelet Transform (DWT) and Discrete Cosine Transform (DCT), and then the SVR train model can be obtained by applying the SVR theory. Finally, the digital watermark can be embedded and extracted by utilizing the strong learning ability of SVR. The digital watermark not only can be correctly extracted under various attacks, but also can locate the part of the audio that has been tampered with and tolerate some incidental processes that have been executed. Experimental results show that our semi-fragile audio watermarking scheme can locate tampered regions without the help from the original watermark and has the advantages such as simple computation complexity, good robustness against shearing attack, and accurate location for tamper.