Research On Image Splicing Detection Based On Markov Feature

With the rapid development of social networks,digital images have become an increasingly important way for people to transfer information to each other because they can convey a large amount of information in a compact manner.However,in recent years,due to the development of various image tampering technologies,people can open and download a lot of image tampering software with perfect functions on the Internet.Many malic io us tamperers use it to tamper with digital images,and this tampering is difficult for people to visually identify and detect directly.Therefore,in order to re-establish the credibility of people's digital images,there is an urgent need to develop a reliable and effective digita l image tampering detection method.In the current research,there are two main methods for image falsification and evidence collection: digital image active forensics method and digital image passive forensics method or digital image blind identification detection method.The digital image active forensics method mainly detects some prior information such as digital watermarks,digital signatures,digital fingerprints,etc.,before the image is generated.If the embedded prior information is found to be modified or destroyed during the detection process,The tamper detection and positioning operations can be performed based on the key and the corrupted informat io n.Compared with the digital image active forensics method,the digital image passive forensics method does not need to add any a priori information when generating the image,and the tampering identification and tampering area is falsified according to the unique tampering marks left in the digital image during tampering.Detection.Therefore,the current research is increasingly approaching the digital image passive forensics method.Image splicing,as one of the most common digital image tampering techniques,is also a tampering method that causes the most serious social impact,and has attracted more and more attention from researchers.Therefore,this paper focuses on the blind identifica t io n detection technology of stitched images.First of al,this paper introduces the research background and significance of array image forensics technology.Then it introduces the active forensics method and the representative method of passive forensics method,and compares and determines the passive forensic technology of stitched images.In passive forensics technology,the inherent labels of different digital images(noise,light ing environment,pixel statistical features,etc.)brought by different operations when generating images.After comparison,this paper selects the statistical features based on image pixels.Passive forensics method.In the previous statistical feature-based detection framework,the detection model often cannot fully utilize the information of all the color channels of the image and the correlation between the channels.Therefore,the algorithm of the third chapter of this paper introduces the concept of "octal number" when extracting the Markov feature of the image for tampering detection,and proposes the DOCT algorithm.The fusion method of highdimensional space transformation two-dimensional space and the parameter selection method of LIBSVM classifier are optimized.The experimental results show that the method proposed in Chapter 3 can effectively improve the performance and robustness of the tampering model.Due to the complexity of DOCT and the loss of information when transforming highdimensional features into two-dimensional space,in the fourth chapter,the tampering model introduces the QDCT algorithm,and in order to completely construct the framework structure in high-dimensional space,QDIFF algorithm and Q-Markov,then introduce the QSVD dimensionality reduction algorithm for the redundancy of Markov features,and finally use QBPNN to classify real images and mosaic images.Experiments have shown that the model exceeds all existing blind detection algorithms for stitched images.