Coverless Multi-Image Steganography for RGB Images: Generative Adversarial Networks based Approach & Steganographic Key based Approach

Abstract

Abstract Inrecentyears,CoverlessImageSteganography(CIS)hasemergedasapromis- ing alternativetotraditionalsteganographictechniquesbyeliminatingtheneed forexplicitdataembeddingandinsteadutilizingnaturallyoccurringorgenerated imagesascarriers.However,currentCISmethodsoftensu!erfromlowhiding capacity,limitedtoconcealingasingleimage,ando!er insu”cientrobustnessand securitywhenhandlingmultiplesecretimagessimultaneously. To addresstheselimitations,wepresentanovelmulti-imagecoverlessstegano- graphic frameworktailoredforRGBimages,capableofsecurelytransmittingand reconstructinguptofoursecretimagesusingasinglefusedcarrierimage.Un- likeconventionalmethodsthatdependonpre-existingcoverimages,ourmodel generates anintermediatecarrierimagethroughasophisticatedpixel-levelfusion techniquethatmergesvisualinformationfrommultipleRGBsecretimageswhile ensuringcompleteperceptualinvisibilityoftheoriginalcontent.Thisfusionpro- cess guaranteesthatthecarrierimagecontainsnovisuallyidentifiabletracesof the concealedimages,o!eringaninherentlysecureandstealthyrepresentation. To furtherstrengthenconcealment,thefusedcarrierundergoesmultipleroundsof imagescramblingtransformations,makingunauthorizeddecodingvirtuallydi”- cult.Wealsointegratearobuststeganographickeymechanism,restrictingaccess and decodingcapabilitiesexclusivelytoauthorizedreceiversequippedwiththe correctkey.Thisensuresthatevenifthecarrierisintercepted,thehiddencontent remainssecureandunrecoverablewithouttheproperdecryptioncontext. In thedecodingphase,weproposearefinementalgorithmdesignedtorestore fine-graineddetailslostduringthefusionandscramblingprocesses.Thisalgo- rithmsignificantlyenhancesreconstructionfidelityandvisualquality,eveninthe presenceofadversarialconditionssuchasnoise,tampering,orgeometrictransfor- mations. ComprehensiveexperimentsonbenchmarkRGBimagedatasetsvalidatetheef- fectivenessofourproposedmodel,demonstratingsubstantiallyhigherhidingca- pacity,minimalvisualdistortion,andresilienceagainststatisticalandperceptual steganalysisattacks.ComparedtoexistingCISapproaches,ourmethodo!ers a moresecure,high-capacity,andvisuallyimperceptiblesolutionformulti-image coverlesssteganography,layingthefoundationfornext-generationcovertcommu- nicationsystems.