Research On Passive-Forensics-Based Detection Method For Homologous/Heterologous Image Splicing

With the rapid development of information technology and the widespread use of multimedia acquisition tools,multimedia data such as image,video,and audio have become the main carrier of information dissemination.However,powerful image editing tools and easy-to-control image processing software make it is ubiquitous to alter the content of digital images without leaving any visible clues.In recent years,the incidents of photo fraud emerge in endlessly.The existence of fake images has lowered the credibility of image information,and has caused serious negative effects in many fields such as politics,economics,scientific research,news media,medical diagnosis,cultural media,judicial forensics,and military application and so on.Therefore,there is a urgent demand for the detection technology of image content authenticity and integrity.This paper mainly focuses on the passive forensics method for image splicing from the same/different sources.The main works are as follows:Aiming at the common forgery attacks of heterogeneous image splicing,we proposed a finer-grained image splicing localization method based on noise level estimation.In the proposed method,we extract statistical feature from the DCT coefficients of image,use such feature to estimate the noise of kurtosis statistic and feature of principal component.On the other hand,we estimate image blocks noise using Laplace operator.The local noise is formed by combining the noise of Laplace operator and the statistical noise of kurtosis.The forensics features are formed by combining the local noise and the principal component feature.Then,we use fuzzy c-means clustering method to cluster the forensics features to divide the image blocks,and the region marking method is used to connect the discrete image blocks into connected regions.According to the area ratio of the image region,the spliced image region is determined.Experimental results show that the proposed method is effective and has higher detection accuracy and better robustness than other methods.Considering that Copy-Move tampering is another common method of image content attack,we proposed an object-based Copy-Move tampering detection method.Different from the detection methods based on image blocks and feature points,the main idea of this method is to introduce the concept of the maximum stable extremum region to the image copy-move tamper detection,and extract the visual objects in the image with the maximum stable extremum region.In this method,we first extracted SIFT feature points that matched each other in the image.Then the maximum stable extremum regions were extracted on R,G and B color channels respectively.The mapping relation between the feature points and the maximum stable extremum region is established and the corresponding maximum stable extremum region is found by matching the feature points.The two channels with the largest number of extremum regions were found,and the color features and sharpness features of the maximum stable extremum regions corresponding to the two channels were extracted,and the suspicious regions were determined by matching.Finally,the region of replication is located through the intersection of the suspicious regions of the two channels.The method has the ability to distinguish the self-similar region and the copy and forgery region in the image,and can detect the copy and forgery region in the image with the self-similar region.Experimental results show that the method has high detection accuracy and strong robustness for content-preserving image processing operations,such as JPEG compression,gaussian noise,salt and pepper noise,median filtering,wiener filtering and gamma correction.