Research On Digital Images Forensics Algorithm

With the development of computer techniques and network techniques, it is easier to tamper and transmit digital images. If tampered images are used in the court, criminal investigation, news media or other momentous occasions by the criminals, it will bring about serious consequences and cause huge losses.Therefore, it is extremely urgent to authenticate the images. Integrity of digital images can be defended by image forensics. The image forensics can be divided into active techniques and passive techniques. The passive techniques can be further divided into image source identification, image tamper detection, and detection of computer generated images. In this paper, the technique of image tamper detection is studied and several algorithms are proposed:(1) FI-SURF algorithm is proposed to detect the tampering of image copy-flip-move. In this algorithm, the arrangement of the descriptors of SURF in the copy-flip-move is studied. In order to detect both the copy-flip-move and the copy-no-flip-move patches in the tampered image, the descriptors of SURF is rearranged. While keeping the advantages of fast response and strong robustness of SURF, FI-SURF makes up the shortfall of SURF in the copy-flip-move tampering detection. The experiments show that FI-SURF is robust to rotation, scaling, JPEG lossy compression, and Gaussian white noise.(2) Rg2NN algorithm is proposed to detect the multiple copy-move. The distance between the feature descriptors is studied in the situation when one patch in the image is copied and pasted multiple times or two or more patches in the image are copied and pasted multiple times. Based on g2NN, Rg2NN is proposed. Rg2NN improves the performance of multiple copy-move detection of g2NN. After feature matching, the feature points are divided into different patches according to the affine transformation between the feature points. The experiments show that the proposed algorithm can extract the feature points in the multiple copy-move tampering images and localize the copy-move patches accurately, thus improves the performance of multiple copy-move detection.(3) An image tamper detection algorithm based on inconsistent perspective geometric is proposed, in which the vanishing points of projection of parallel lines are utilized to authenticate the images. After edge detection and straight lines extraction,the intersection points of the parallel lines are computed, and the normalized mean value and the normalized standard deviation of the distances between the intersection points are used as the evidence of image tampering. Based on the understanding of the image content,parallel lines which are suitable for detection are extracted. In the images which contain two or more vanishing points, in order to compute the intersection points of the parallel lines, the parallel lines with the same direction should be extracted one time, and the parallel lines with another direction should be extracted the next time. The experiments show that the proposed algorithm can be used to authenticate the images,and the detection performance is improved compared with the algorithm of non-normalized mean value and the non-normalized standard deviation.