Image Information Hiding Detection For HUGO Algorithm

With the rapid development of Internet and communication technology, especially the extensive development and application of digital watermarking and steganography, makes the hidden information can much easier secret communication in the channel.At the same time,those technology bring convenience to people,also easy to be used by the criminals.Steganography and digital watermarking abuse will cause seriously damage to the stability of society and affect people’s lives.Information hiding technology is a important way to prevent the spread of illegal information.Its main task is to detect the existence of any confidential information,estimate the length of secret message,even estimate that the key to crack steganographic algorithms. Image possesses many characteristics,such as redundancy,extensive usability,it very suitable for the carrier of secret communication.Therefore,research of image information hiding detection technology has important theory value and practical significance.HUGO steganography is a kind of hard to detect steganographic technology, do research for HUGO has the certain difficulty and challenge.Based on the analysis of digital steganography and steganalysis, the theory of understanding of relevant concepts and models, technical indicators, in order to further improve the detection effect, two kinds of detection algorithm for HUGO is proposed by us and do experiment verified.This paper main research work are as follows:(1) Proposed based on co-occurrence matrix of residual HUGO image steganography detection.First calculating higher order residuals, and then used residuals calculated co-occurrence matrix looking for more differences characteristics,finally select classifier for experimental verification.At the same time, the experimental results show that the proposed algorithm has very good detection rate on HUGO,EA, LSB matching steganograpy algorithm.Further extracting SQUARE, MINMAX, KB, MARKOV and EDGE features do contrast experiments,it is prove that SQUARE and MINMAX under the condition of same detection rate is higher than other types of features.Finally when choosing classifier training,we contrast Ensemble Classifier and Support Vector Machine classifier in the same conditions of detection rate and time complexity, experiments show that the Ensemble Classifier is the best choice.(2) Proposed based on Local Linear Transformation HUGO image steganog- raphy detection.Using local linear Laws template to convolution the image which will be detected high-pass filter for texture image, then extract the co-occurrence matrix(span pixel area=3,5) as its features, choose Ensemble Classifier to classify the training Set and testing Set,got an average of82.71%detection accuracy rate in BOSSrank image Library.In order to do not break in general, we random sampling7500image from BOSSbaseV0.92,BOWS2,NRCS,UCID four images Library make up self-built image Library experiment,Then We contrast of different images Library detection error rate in different embedding rate effect,drawing Receiver Operating Characteristics Curve.Finally we obtained the average detection error rate Pe of self-built image Library.