面向隐写失配的MRS图像隐写分析模型OA

An adaptive steganalysis model that employs multi-representation structure(MRS)and conditional maximum mean discrepancy is proposed to address the challenge of diminished detection accuracy caused by cover source mismatch in the context of un-known steganography algorithms within modern image steganalysis deep networks.On one hand,and MRS is constructed to capture fea-ture distributions from various structures and acquire multiple domain-invariant representations.On the other hand,conditional maximum mean discrepancy(CMMD)is employed to minimize differences in feature distributions while concurrently aligning global feature distri-bution and the distributed sub-domain classification features.The cross-mismatch experimental results for three representative steganog-raphy algorithms demonstrate the effectiveness of the proposed method in effectively mitigating the issue of diminished detection accura-cy attributed to cover source mismatch.In comparison to recent conventional methods,an improvement of up to 10.47%is observed at an embedding rate of 0.2 bit/pixel.