Research Of Steganalysis Scheme Based On Local Texture Features

Steganography is a technique that hides information into digital cover images.The main characteristics of steganography is its invisibility, that is, other people arenot aware of the hidden information in the image. And steganalysis algorithm tries totell whether there is hidden data in an image according to various known information.Attacking HUGO algorithm is one of the research focus in the area of steganalysis.HUGOisonekindofcontent-adaptivespatialdomainsteganographicalgorithm,whichuseshighdimensionSPAMfeaturestoanalysisthecostofembeddingbitsintodiferentlocations of the cover image, picks out some pixels and alters the values. Thus, it canresists attacks from most known steganalysis methods.This paper described and analyzed the advantages of three existing steganalysisschemes targeting HUGO, taken the particularity of HUGO steganography algorithminto account and proposed a steganalysis scheme based on improved local texture fea-tures. The proposed approach extracted LOCP and LPQ features from original in-put image, diference images and prediction residual images and combined these fea-tures. As a result of high dimension of these features, Ensemble classifer was usedfor training and testing, and proper method was also introduced for dimension reduc-tion. The experiment on BossBase1.0image database demonstrated that, the proposedschemeperformedbetterindistinguishingstegoimagesandcoverimages,andachieved84.65%in detection accuracy using our30658-D feature sets.