Digital image forensics technology research
|School||Zhejiang University of Technology|
|Course||Signal and Information Processing|
|Keywords||Forensics CFA BP Network DCT Dictionary sort|
Thanks to LSI and advances in information technology, the high cost of digital cameras such as the original old Yan Wang Xietang ago, flying into the homes of ordinary people. Now, in our daily life and work, digital cameras have almost completely replaced the traditional film cameras become the most important tool for image recording. However, the photos taken with digital cameras have inherent digital features and storage features, so photographs taken with conventional cameras, the digital images are more likely to be tampered with. So, how can the use of digital photos before determining whether the digital photo has been tampered with, has become news, economic, military, political and other fields admissible photo before must face. The traditional approach is embedded in the digital image digital watermarking, as long as tampering tampering with the image, then surely damage the image embedded watermark, which can determine whether the image has been tampered with. However, the watermark embedded in the image destruction of the visual effect of the image, is an imperfect means of defense. If you can not, under the premise watermarked image statistics and other means to determine through a digital image has been tampered with before, namely, \The problem studied in this paper belong to this category. This study can be divided into two parts, one is proposed \\The first part, based on the principle of digital camera imaging proposed \That the use of the image itself carried at the time of imaging digital camera interpolation algorithm information to determine whether an image has been tampered with. If the image interpolation algorithm contains two or more, then this image can be determined is tampered with, the image included in the area with the most different interpolation algorithms tampered area is a small area. To achieve the above object, this use of the original BP neural network training images, the image of the interpolation function fitting, and then the color image information input fitting interpolation function, the fit function obtained by the color information of the image and the actual image color information to determine the size of the error forgery area. In order to further improve the accuracy of forensics, this paper further proposes to use the weighted value of the size selection whole image interpolation between the best-preserved part to train the BP neural network, it also improves the algorithm for salt and pepper noise robustness. The second part, the first part of this paper, the proposed algorithm is for forensics from different photos taken with the camera image spliced into a situation by the line forensics. In practical applications could also use data from tampering with a camera captured images together into a fake image. To deal with this tampering, predecessors have made some forensics algorithm, based on image block eigenvector and dictionary sorting algorithm for digital image forensics is one of them. The actual tampering image blurring due process, similar to the image block dictionary sort may not be adjacent, and therefore will lead to detection errors. This paper, the algorithm is proposed to improve the distance between the feature vectors according to the size of the dictionary sort, and then follow the feature vector frequency DCT coefficient matrix method for grouping, thereby improving the accuracy of the algorithm.