Effective Medical Image Copy-Move Forgery Localization Based on Texture Descriptor

Medical images are vulnerable to be maliciously tampered during network transmission, affecting diagnosis of doctors. Moreover, some images in medical research papers are intentionally manipulated, which reduce the credibility of the conclusions. Therefore, it is essential to research an effective and robust algorithm for medical image tamper detection ans localization. In this paper, we propose a copy-move forgery localization algorithm for medical images called MITD-CMFL. Due to texture structure information is complex and important for medical images, we obtain textural images from noise-reduced images by utilizing texture descriptor to gain more accurate features. It is difficult to extract a sufficient number of feature points with strong representation ability in smooth regions to characterize textures, we extract SIFT keypoints in texture images and decrease the contrast threshold. The experiments conducted on 2,898 tampered breast cancer images randomly selected from DDSM dataset show the pixel-level \(F_1\) of MITD-CMFL reaches up to 95.07% under plain copy-move attack, and the method has superior performance even under typical image transformations compared to the state-of-the-art algorithms.