Copy-move detection of digital audio based on multi-feature decision

Copy-move forgery is one of the common means of audio splicing. With the advantages of convenience and covertness, copy-move forgery is widely used in our daily life, which is difficult to be found out. This paper proposes a multi-feature decision method to detect copy-move forgery. We use C4.5 decision tree to merge together the detection result of four features which are gammatone feature, Mel-frequency cepstral coefficients (MFCCs) feature, pitch feature and discrete fourier transform (DFT) coefficients. To enhance the detection effect, we use Pearson correlation coefficients (PCCs) and average difference (AD) to evaluate the detection result of these four features. Compared with other single-feature decision methods, the proposed scheme has better detection result of copy-move forgery. The experiment shows that the proposed scheme is effective and robust with the data of 1000 normal audio and 1000 copy-move audio. (C) 2018 Elsevier Ltd. All rights reserved.