Two approaches for integrating encryption with multimedia compression systems are studied in this research, i.e., selective encryption and modified entropy coders with multiple statistical models. First, we examine the limitations of selective encryption using cryptanalysis, and provide examples that use selective encryption successfully. Two rules to determine whether selective encryption is suitable for a compression system are concluded. Next, we propose another approach that turns entropy coders into encryption ciphers using multiple statistical models. Two specific encryption schemes are obtained by applying this approach to the Huffman coder and the QM coder. It is shown that security is achieved without sacrificing the compression performance and the computational speed. This modified entropy coding methodology can be applied to most modern compressed audio/video such as MPEG audio, MPEG video, and JPEG/JPEG2000 images.

A fragile speech watermarking scheme of low complexity is proposed in this research as an effective way to detect malicious content alterations while tolerating content preserving operations. Since no auxiliary data are required, the authentication process is transparent to agents that perform content preserving operations such as transcoding during the transmission process. The proposed scheme is based on the modified odd/even modulation scheme with exponential scale quantization and a localized frequency masking model while assuring no mismatch between quantization steps used in watermark embedding and detection. The noise introduced by watermarking is shown to be smaller than that of speech coders. It is shown by experimental results that the proposed system is able to distinguish malicious alterations from resampling, white noise pollution, G.711 speech coding and G.721 speech coding with very low error probabilities.

A speech content authentication scheme, which is integrated with CELP speech coders to minimize the total computational cost is proposed in this research. Speech features relevant to semantic meaning are extracted, encrypted and attached as the header information. A low cost synchronization algorithm is used to resolve mis-synchronization caused by content preserving operations. Silent and tonal regions in the speech are identified with algorithms of low complexity to enhance the precision of integrity verification. This scheme is not only much faster than traditional cryptographic bitstream integrity algorithms, but also more compatible for a variety of applications. Experimental results are collected by using the GSM-AMR speech coder, and statistical analysis is performed to calculate the false positive rate of tamper detection.