Motion vector reversion-based steganalysis revisited
Peipei Wang, Yun Cao, Xianfeng Zhao, Bin Wu · 2015
In this paper, we revisit a video steganalytic scheme using the motion vector reversion-based (MVRB) feature, and make necessary improvements. Since been proposed, MVRB feature has shown its effectiveness in detecting typical motion vector-based steganographic schemes. However, as also noticed by other studies, the effectiveness of MVRB feature highly relies on the re-compression operation. If the parameters used in re-compression differ much from the prior compression, the feature's effectiveness will deteriorate severely. To address this issue, efforts are made to recreate the prior compression context. First, the useful parameters that can be directly obtained are identified and gathered during de-compression. Second, the prior used motion estimation method is inferred using a small range exhaustive matching method. The experimental results have shown that, in detecting MPEG compressed videos, the improved feature does not suffer performance loss any more.