Blind Forensics Of Resampling In Digital Images

As we enter the digital age, people no longer believe what they see in an image because of image tampering, which causes a serious trust crisis. To restore the social trust system, digital image forensic is becoming a hot research topic. This dissertation introduces digital image forensics, and elaborates the blind forensics especially. Our research focuses on the resampling detection which is often used to cover the visual traces of images splicing. Inspired by prior works, we propose two new algorithms for detecting resampling in digital images.1. Detection of resampling based on correlation coefficients. We observe that the correlation coefficients of adjacent rows(or columns) vary periodically in resampled images and the period is relevant to the resampling rates. So the correlation coefficients can be used to detect resampling and estimate the resampling rates. As experiments show, our algorithm is more robust than prior works and shows better detection results.2. Detection of resampling based on redundancy. Current resampling detection methods cannot determine the resampling ratios accurately. The reason is that the resampling ratio is estimated using the position of peaks in frequency domain, but due to frequency aliasing, the position of peaks and the resampling ratio are not in one-to-one correspondence. To solve this problem, we propose an algorithm which can determine the rasampling ratios in spatial domain. First we introduce the meaning of redundancy in digital images and how to measure the redundancy, then we point out that the redundancy of the rows(or columns) of a resampled images vary periodically and can be used to estimate the resampling ratios. As experiments show, our algorithm can correctly detect all resampling ratios in uncompressed images, and in compressed images the detection results are better than prior works.