LSB steganalysis based on high-order statistics

We propose a general framework for the detection of the LSB steganography using digital media files as cover objects. The new framework exploits high-order statistics of the samples. It can compute a robust estimate of the length of a secret message hidden in the least significant bits of samples for a large class of digital media contents such as image, video and audio, in which the underlying signals consist of correlated samples. A case study on the LSB steganalysis of natural grey-scale and color images and experimental results are reported.