Research On Fine-grained Image Splicing/Composition Detection Method

With the rapid development of digital technology,and the wide use of various powerful image processing tools,non-professionals can beautify,edit,even modify and forge digital images without leaving any visible clues,which will destroy the primitiveness,integrity and authenticity of image content.Meantime,the existence and dissemination of false images reduces the credibility of digital content,and causes serious negative effects in many fields such as scientific research,news media,judicial forensics,finance and military and so on.Therefore,it is urgent to develop powerful image tampering detection tools/algorithms to identify the tampering of image content and ensure the primitiveness,integrity and authenticity of image content.In this thesis,we devote to study the detection technology for image splicing/composite forgery,the main works are as follows:We propose a coarse-to-fine grained image splicing region detection method.Most digital cameras have a single charge coupled device or complementary metal oxide semiconductor sensor,and through Color Filter Array(CFA)to get a color image.According to the fact that image splicing operation may destroy the consistent linear correlation pattern introduced by CFA interpolation,we can locate the image spliced region by estimating the local CFA interpolation pattern of the image.In the method,firstly,we use the covariance matrix to reconstruct the R,G and B three color channels of the image,so as to estimate the CFA interpolation pattern used in the image.Then,according to the difference between the estimated CFA interpolation pattern and the original CFA interpolation pattern to construct the image forensics features,so as to perform the coarse-grained image splicing region detection to obtain the suspicious spliced region.Later,we use singular value decomposition technique to extract local texture strength features from coarse-grained detection result,and perform the fine-grained image splicing region detection by classifing the features.Finally,we use the super pixel segmentation algorithm to smooth the edge of the fine-grained splicing region,so as to remove the false detection,and obtain the accurate splicing region detection result.Compared with the current detection methods,the proposed method shows high detection accuracy,low time complexity,strong generalization and robustness,solves the problem that CFA-based image splicing region detection methods are not robust to JPEG compression operation.We propose a fine-grained image splicing region detection method based on CFA features.Considering that the spliced region and the original region may have different CFA interpolation pattern and different pattern noise level in the spliced image,we use the difference as evidence of image splicing tampering to detect image spliced region.In the proposed method,firstly,we use Expectation Maximization(EM)algorithm to estimate the difference of CFA interpolation pattern,define the difference as CFA noise,apply wavelet transform to estimate image noise,and predict the local weighted noise variances of the two kinds of noise images.Then,we define and construct image CFA features according to the estimated local weighted noise variances.Later,apply the fuzzy C-means clustering algorithm to cluster the features to obtain suspicious image spliced region.Finally,we use Canny operator to remove the false detection in the suspicious splicing region detection result to obtain the final image splicing region detection result.Compared with the existing related methods,the proposed method is provided with satisfactory performance,it can accurately detect the position and shape of the image spliced region,and has better robustness for the common image content preserving operations.