Image Compression Tamper Detection Based On Multi-feature Fusion

With the advancement and development of science and technology,digital image editing and processing tools are also becoming more and more advanced.Digital images are often tampered for different purposes by some people,which affect the society in varying degrees.At the same time,it also leads to the public crisis of trust in digital images.As one of the most widely used image format at present stage,JPEG format images seen in daily life are the most sources of image tampering.Therefore,forgery detection of JPEG images has very important academic significance and practical value.Its strong development trend and wide range of applications have attracted great attention of many scholars,and have become the focus of research in the international academic community.In this thesis,the statistical characteristics of double JEPG compression is considered for forgery detection and tampered region location,wherein a method based on the first digit feature extraction of DCT(Discrete Cosine Transform)coefficients is presented,and a tampering detection algorithm based on the feature fusion is proposed.The main research work of this thesis is as follows:Firstly,the JPEG related compression technique is described in detail,and the principle of JPEG compression,decompression and recompression are introduced respectively.Some typical image tampering operations and the general way of image tampering are discussed.Meanwhile,the image tampering localization model used in this thesis is also given.Secondly,considering that the first digit feature of DCT coefficients has become one of the hotspots for detecting JPEG image recompression in recent years,an improved forgery detection and tampering localization method is proposed for double JPEG compression images based on first digit feature extraction of DCT coefficients.This method adopts PCA(Principal Component Analysis)dimensionality reduction to extract the compact features of the first digit feature of DCT coefficients,and then selects an appropriate kernel-based nonlinear classifier called KNR(Kernel-based Nonlinear Representor)to detect JPEG recompression operations and locate the tampered regions.Experimental results show that in comparison with representative algorithms,the improved algorithm achieves better results,and is robust to operations such as rotation,resizing and feathering.Moreover,KNR classifier outperforms classical SVM(Support Vector Machine)classifier in recognition effectiveness and efficiency.Finally,in order to solve the problem of forgery detection of JPEG images better,an image compression tampering detection method based on feature fusion is proposed.This method utilizes the classical DCT coefficient histogram feature and the first digit feature of DCT coefficients for feature fusion.By using different features,it overcomes the defects of JPEG tampering detection based on a single feature.Experimental results show the effectiveness of multi-feature fusion for forgery detection of JPEG images.