Research Of Blind Image Tamper Detection Based On Hypercomplex Transform Domain

In recent years, with the rapid development of mobile digital technology and multimedia Internet, people can easily make use of mobile phones, digital cameras and other multimedia devices to shoot their favorite pictures and videos and use the software edit them freely, then these are uploaded to the Internet. It is difficult to protect the sources, integrity, authenticity of digital image and video data, which not only violate the copyright of legitimate digital video images, it is but also difficult to serve as an effective basis for judicial decisions. Hence, whether or not it is valid for identification and authenticity of video images become a hot topic of people’s attention. Our works are as follows*:1) For Image splicing tamper detection algorithms can not sufficient use color information of color images and high complex modeling in image splicing tampering detection methods, therefore an image splicing image tamper classification detection algorithm based on Markov features in QDCT(Quaternion discrete cosine transform) domain is proposes in this paper. This method can reduce modeling complex and overcome insufficient use color information of color images. First of all, the input color image is divided into 8?8 non-overlapping blocks, color information of each block’s RGB three-channel are constructed quaternion block image, then QDCT quaternion transformation is processed to quaternion block image of each block. Further, QDCT domain extension Markov features of each image are calculated. Finally, SVM(Support Vector Machine) is used to classify the natural images and forgery images. Compared with the existing method in same database, the proposed algorithm obtain a higher classification accuracy and is better than other algorithms.2) For Copy-move tamper detection algorithm robust reprocessing poor, high redundancy algorithm problems, therefore this paper proposes an image Copy-move tamper detection method based on QDCT domain. This method can take advantage of the characteristics of color information and reduce redundancy. Firstly, the input color image is divided into 8?8 overlapping blocks. Then RGB three-channel 8?8 overlapping blocks QDCT are done for the color images which are to be detected, the limited QDCT coefficients are extracted after scanning by Zig-zag, and they are as a feature vector of each block. then characteristic matrix is sorted by dictionary sort according to row. Finally, euclidean distance of the target component and the adjacent component are calculated for feature matrix after traverse and sort so as to find the minimum distance. When the minimum distance is less than a predetermined threshold value, the position of two blocks are recorded, and a copy-move forgery region is located. Final results show that this algorithm can locate the tampered region, and compare with other algorithms our method are more robust in post-processing.3) There are some problems in traditional video frame insertion and deletion tampering detection methods in video tampering, these problems has high complexity in modeling and can not detect tampering quickly and efficiently. Therefore an blind detection algorithm of video inter-frame tampering based on QWT(Quaternion wavelet transform) domain perceptual hash is put forward in this paper, which can reduce modeling complexity quickly and extract tamper type feature tamper type efficiently. Firstly, the eigen values of QWT domain perceptual hash between adjacent two frames are calculated. Then the number of outliers are detected for extracted feature vector by threshold determination. Finally, according to the number of outliers to achieve the detection and localization in video frame insertion and deletion tampering. The algorithm can detect and locate the video frame insertion and deletion tampering accurately, at the same time, it can better distinguish the frame insertion tampering and frame deletion tampering efficiently.