Steganalysis Embedding Percentage Determination withLearning
Kenneth W. Bauer · 2006
Steganography (stego) isusedprimarily whenthe veryexistence ofacommunication signal istobekeptcovert. Detecting thepresence ofstego isaverydifficult problem which ismadeevenmoredifficult whentheembedding technique is notknown. Thisarticle presents aninvestigation oftheprocess andnecessary considerations inherent inthedevelopment ofa newmethodapplied forthedetection ofhidden datawithin digital images. We demonstrate theeffectiveness ofLearning VectorQuantization (LVQ)asa clustering technique which assists indiscerning clean ornon-stego images fromanomalous orstego images. Thiscomparison isconducted using 7features (1) overasmall setof200observations withvarying levels of embedded information from1% to10%inincrements of1%. Theresults demonstrate thatLVQ notonlymoreaccurately identify whenanimagecontains LSBhidden information when compared tok-means orusing just therawfeature sets, butalso provides asimple methodfordetermining thepercentage of embedding given lowinformation embedding percentages. I.INTRODUCTION