Research On Key Technologies Of Digital Image Authenticity Identification

With the rapid development of digital image processing technology,a large number of efficient and intelligent digital image processing software has emerged,which makes image editing very easy,and will not leave any traces of human visual perception.However,digital image processing technology is a "double-edged sword",which not only meets the needs of people for images beautification and retouching,and other normal image modification,but also becomes a tool for a small number of organizations or individuals with "ulterior motives" to tamper with images.Once the forged digital image is widely spread through the Internet and other mass media,it will have a very bad impact on the country,society and individuals.Therefore,how to effectively and accurately identify the authenticity of digital images has become the research topic of this paper.The reason why tampering with digital images is disgusting is that it transmits false and wrong semantic information.Therefore,how to effectively detect the semantic level tampering of digital image is the key problem in the field of digital image authenticity identification,and the key technology is also the focus of this paper.Common semantic image tampering technologies include content aware image resizing(CAIR)and exemplar-based image inpainting.With the help of digital image forensics technology and methods,this paper studies the key technologies and methods to effectively identify the above two semantic level image tampering.In order to enhance the image performance,the normal image processing should be affirmed,but it is the malicious tampering that arbitrarily changes the image content.Based on the existing theories and methods of digital image forensics,this paper proposes a new model of digital image authenticity identification.This model distinguishes the normal modification and malicious tampering,analyzes and explains the image processing in more depth and detail,and then judges whether the image processing affects the authenticity of the image.The main work and innovation of this paper include:(1)Aiming at the image tampering of seam carving,an image tampering detection method based on LNMOP feature and HOG feature is proposed.Through the analysis,it is found that the seam carving can change the magnitude distribution of the pixel intensity difference in the local neighborhood.In order to describe the change accurately,a novel image pattern description method,Local Neighborhood Magnitude Co-occurrence Pattern(LNMOP),is proposed.LNMOP is different from traditional descriptors(such as LBP),and has strong ability of magnitude recognition and anti-noise.In this paper,the detection method of image seam carving is proposed.Firstly,LNMOP and HOG features are extracted from the image;then,the best features for classifier are selected from the extracted LNMOP features;finally,support vector machine(SVM)is used to train and test the best features to distinguish tampered image and normal image.Aiming at the problem of screening the best features of LNMOP,this paper proposes a method of LNMOP feature selection based on the hierarchical matching of HOG features,which effectively reduces the dimension of LNMOP features and the computation of classifier.This method determines the best LNMOP features to be screened according to the hierarchical level of HOG features.The experimental results show that the proposed method not only overcomes the insensitivity of traditional Markov features to small ratio tamper detection,but also avoids the problem of LBP features confusing texture region and smooth region due to noise.Compared with the current representative methods,the detection performance is improved.(2)Aiming at image tampering based on content aware image resizing technology,a detection method based on improved local ternary patterns(ILTP)and gradient energy feature(GEF)for image tampering is proposed.It is found that content aware image resizing(CAIR)can destroy the correlation of local neighborhood pixels.Although the local binary patterns(LBP)can describe the correlation to some extent,it cannot describe the magnitude information of local neighborhood pixels and is easy to be interfered by noise.For this reason,the local ternary patterns(LTP)operator is improved in this paper,which makes the threshold t have strong adaptive ability and recognize the change of local texture more accurately.The CAIR operation of image will inevitably lead to the change of energy distribution in the image,especially the energy accumulation in the gradient direction.In this paper,a new feature,gradient energy feature(GEF),is proposed to describe the energy distribution in the gradient direction.In this paper,an image tamper detection method based on ILTP and GEF is proposed,which makes use of the above two features.Firstly,ILTP and GEF features are extracted from the candidate images and connected into combined features.Then,the classifier is trained with combined features.Finally,two groups of experiments are designed to verify the important role of ILTP and GEF features in CAIR tamper detection.Compared with the existing methods,the detection accuracy is greatly improved.(3)Aiming at the object removal based on exemplar-based image inpainting technology,this paper proposes a detection method based on LSTM-CNN for image object removal.Through the research and analysis,it is found that the existing exemplar-based image inpainting forensics methods generally have the following disadvantages:the abnormal similar blocks are time-consuming and inaccurate to search,high false alarm rate and lack of robustness to post-processing combination operation.In view of the above shortcomings,a detection method based on LSTM-CNN for image object removal is proposed.In this method,convolutional neural network(CNN)is used to search for abnormal similar blocks.Because of CNN's strong learning ability,it improves the speed and accuracy of search.LSTM network is used to eliminate the influence of false alarm blocks on detection results and reduce the false alarm rate.A filtering module is designed to eliminate the attack of post-processing operation.Experimental results show that the method has a certain accuracy,and can resist the attack of post-processing combination operation.(4)In order to distinguish the normal modification and malicious tampering of digital image,a novel identification model of digital image authenticity based on life feature is proposed.The model starts from the life feature of digital image,interprets and evaluates the detected processing combined with the specific life feature,and takes the difference of image authenticity before and after processing as the evaluation standard,through establishing the authenticity evaluation model.The model can synthetically judge whether the digital image is authentic or not,so as to avoid the false judgment of the authenticity.Finally,the validity of the proposed model is verified by experiments.To sum up,for semantic image tampering,this paper proposes a series of key technologies from three aspects:the forgery detection technology for seam carving image,the forgery detection technology for CAIR image and the object removal detection technology for exemplar-based image inpainting,so that the forged digital image cannot hide in front of these powerful technologies.Finally,based on the semantic level digital image tampering identification technology,this paper proposes a life feature based authenticity identification model for digital image,which can effectively distinguish the normal modification and malicious tampering of the image,improve the accuracy of digital image authenticity identification,and expand the development space for the semantic level digital image tampering identification.