Analysis Methods And Detection Techniques For Double Compressed Audio

The development of digital multimedia communication technology, enrich the way peoplecommunicate and spice up our lives. However accompanied by plenty of digital multimedia editingsoftware, these tools make us modify these digital media informations easily. These intentional orunintentional tampering, make the public be vigilant to the reliability and integrity of multimediainformations. With the increasing demands of the digital information on copyright protection and courtforensic internationally, the digital multimedia forensics has become a hot research topic. As one of theimportant branches of multimedia forensics, the development of the digital audio forensics falls behind theimages and video forensics. But its major practical significance can not be ignored.Digital Audio compression detection is an important part of audio forensics, such as cutting, splicingand other tampering operations will cause double audio compression. As popular music formats on theinternet, MP3and AAC are focused in this research subject. This paper starts from the internal audiocoding theory, and aims for the detection algorithm of the double MP3compressed audio, also makestentative exploration on AAC audio format.The main research achievements achieved in this thesis include the following three aspects.Firstly, based on the non-uniform quantization process of the MP3encoding, this paper analyzes thethree nested loop in quantification, and proposes an algorithm for the double MP3compression detectionbase on scale factor difference. Due to the scale factors of the normal audio are greater than that in doublecompressed audio under the long-window encoding, the mean difference of corresponding frame is treatedas the effective features to determine whether an audio has been suffered the secondary compression.Secondly, according to in-depth research and analysis on the Benford’s law which is traditionallyapplied on financial forgery analusis, this topic applied the rule on the quantified MDCT coefficients in theprocedure of the MP3encoding. By experimental analysis, the Benford’s curve of the normal audio have acertain degree of deviation compared with the standard curve, while the curve of the double compressedaudio appears serious distortion. Therefore, these two types of features can be classified by Support VectorMachine.Thirdly, this paper presents a novel scheme which is that the selection of the Huffman tables can betterreflected the diversification of the MDCT coefficients on the other side. Therefore, the core of thealgorithm is based on Huffman table indexes. The Huffman code table selection is substantially similarunder one same bitrate, while the difference is greater under different bitrates. Therefore, this issue putsforward a new method for double MP3compression detection based on Huffman table indexes, and makesattempts on the AAC audio format.