传感技术学报2026，Vol.39Issue(3)：528-535,8.DOI:10.3969/j.issn.1004-1699.2026.03.009
面向隐写失配的MRS图像隐写分析模型
An Image Steganalysis Model Based on MRS for Steganographic Mismatch
摘要
Abstract
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.关键词
图像隐写分析/载体源失配/多表征结构(MRS)/条件平均最大差异(CMMD)Key words
image steganalysis/cover source mismatch/multi-representation structure(MRS)/conditional maximum mean discrepancy(CMMD)引用本文复制引用
叶学义,陈海颖,薛智权,王佳欣,应娜..面向隐写失配的MRS图像隐写分析模型[J].传感技术学报,2026,39(3):528-535,8.基金项目
国家自然科学基金项目(U19B2016,60802047) （U19B2016,60802047）
