Research On Image Copy-paste Forgery Forensics Based On Local Color Invariant Features

With the rapid development of the internet and the popularization of the digital imaging equipments,digital images have become the main source of information.The development of the digital image processing technology has brought about the problem of image tampering,which has seriously affected the credibility of digital images as the event records.How to determine the authenticity of an image has become an urgent problem.Nowadays,the digital image forensics technology has become a new research field,which aims at revealing digital image tampering operations.One of the most common tampering operations is copy-paste,which is often accompanied by a variety of affine transforms,illumination change and degradation operations.There are still some defects in existing image copy-paste forgery forensic algorithms,such as weak robustness against some tampering operations,high computational complexity or inability to identify smooth regions.It is worth noting that,because the color feature is very easy to be influenced by the illumination change,the image color information is seldom considered in existing algorithms.However,color is one of the most intuitive features of objects.It can enhance the discriminative power of objects.In view of these problems,this dissertation studies the local color invariant feature and its application in image copy-paste forgery forensics.The main contributions are summarized as follows:1.Two image copy-paste forgery forensic algorithms are proposed based on the local color invariant features Opponent SIFT and Cw SURF separately.The affine transformations between the matched feature descriptors are estimated before the algorithm of Random Sample Consensus being performed to remove the false matches.Experimental results show that these two algorithms have solved the problem of traditional local invariant feature based methods being unable to identify the flat regions.Moreover,they have certain invariance to many kinds of affine transformations,illumination change and degradation operations.2.A new local color invariant feature OwSURF is proposed by merging the geometrical and color information.And an image copy-paste forgery forensic algorithm is proposed based on OwSURF.The hierarchical agglomerative clustering algorithm is applied to optimize the process of feature matching.Experimental results show that this method is much more robust against various postprocessing tampering operations,while its computational complexity is relatively lower.3.An image copy-paste forgery forensic algorithm is proposed based on local col-or invariance feature Color-SIFT.A framework of extracting Color-SIFT is constructed.Five color invariants are extracted under different imaging conditions as well as the five corresponding local color invariant feature descriptors: Ew SIFT,Ww SIFT,Cw SIFT,Nw SIFT and Hw SIFT.We evaluated the performance of several traditional algorithms with the proposed algorithms in this dissertation.Experimental results show that Ew SIFT and Ww SIFT have stronger robustness to various postprocessing operations.Ew SIFT and Ww SIFT are more suitable for the practical application.