A Novel Mapping Scheme for Steganalysis

摘要

Recently the research on steganalysis for breaking HUGO has been further moved ahead. A novel mapping scheme is reported in this paper. Through a Huffman coding like procedure, this scheme can lower the feature dimensionality from 625 to 120 generated from one residual image as a 4th order co-occurrence matrix is considered. Two experiments have been reported to demonstrate its effectiveness. In breaking the HUGO, the proposed mapping scheme has been applied to the frame work of the state-of-the-art [13] with some minor modification. With a total number of 15,840 features the new method can achieve 87.17% accuracy in BOSSbase ver. 0.92 at 0.4 bpp, outperforming the state-of-the-art.