Research On Detecting Technology For Copy-Move Image Forgery Under Trace Concealment

In recent years,with the increase of digital image forgery events,problem of network information security become increasingly prominent.Copy-move forgery is a common and easy-to-implement image forgery,where one segment of image is copied and pasted in the other part of the same image,making a little change to the real image and being not easy to be noticed.The existing research on copy-move forgery detection suffer from following limitations:(1)Image modification in local subtle area makes copy-move forgery detection more difficult.(2)In order to hide the traces left by copy-move forgery,various transformations such as noise adding,image blurring and part deformation are often applied to the tampered image.In our work,we apply deep learning into copy-move forgery detection to address the limitations mentioned above.The main contributions are as follows:(1)Focusing on the poor performance of the existing methods under small-region tampering samples,the copy-move forgery detection based on self-correlation pyramid network is proposed.The algorithm firstly uses VGG16 architecture for feature extraction,and then selects multiple feature maps of different dimensions to measure feature similarity and construct self-correlation pyramid,enhancing forgery signal in small regions by fusing the similarity calculation results of global feature and local feature,and finally generate a copy-move mask.Detailed results and analysis from experiments suggest that the proposed method outperforms the state-of-the-art algorithm,especially achieves excellent performance under small-region tampering samples.(2)Aiming at the problem of trace concealment in tampered images,a copy-move forgery detection method under trace concealment is proposed.We design a multi-task branch architecture to resist various post-processing attacks by increasing a restoration network for forged image reconstruction.Robustness evaluation is conducted under various attacks,including JPEG compression,noise adding,image blurring,brightness change,color reduction and contrast adjustments.From the numeric results,it was proved that the robustness of the presented algorithm is quite well.