Research On Audio Copy-move Forgery Detection And Localization

With the wide application of audio editing software,audio files can be maliciously tampered for unlawful purposes,which bring serious challenges to the audio authenticity detection.Copy-move forgery is the most common method in audio semantic forgery.The properties of the tampered segment and the original one is very similar,and it is difficult to detect and locate the tampered audio.Therefore,audio copymove forgery detection and localization has become one of the key topics in the field of multimedia forensics.This dissertation focuses on the research of audio copy-move forgery detection and localization.The main works are as follows:(1)Robust audio copy-move forgery localization(CMFL)approach is proposed using sliding window strategy.Specifically,the audio recording is first segmented into voiced segments and unvoiced segments using spectrum entropy based Voice Activity Detection(VAD)algorithm.Then,a sliding window strategy(SWS)is presented to further partition each voiced segment into overlapping small slices.Afterwards,algorithms for similarity computation for two small slices are proposed based on Constant Q Cepstral Coefficients(CQCC)and Pearson correlation coefficients(PCCs).Finally,the integrated method,named SWS-CQCC,is evaluated against the state-ofthe-art copy-move forgery localization on Librispeech and Chinespeech,respectively.The experimental results demonstrate that the SWS-CQCC exhibits significantly effectiveness with respect to different types of copy-move forgeries,and posses high robustness against signal processing attacks.(2)A robust copy-move forgery detection and localization method based on Constant Q Spectral Sketches(CQSS)and Genetic Algorithm is proposed.Firstly,the CQSS feature is extracted by averaging the logarithm of the squared-magnitude Constant Q transform,and the effectiveness of the CQSS is analyzed;Secondly,by designing coding method,genetic operator and fitness function,CQSS optimization method based on Genetic Algorithm is proposed to remove redundancy and interference information in CQSS feature and further improve detection performance.The experiment results show that the proposed method not only has more effective detection and localization performance of copy-move forgery,but also has good robustness against signal processing attacks.