Recapture Image Forensics Based on Laplacian Convolutional Neural Networks

Recapture image forensics has drawn much attention in public security forensics. Although some algorithms have been proposed to deal with it, there is still great challenge for small-size images. In this paper, we propose a generalized model for small-size recapture image forensics based on Laplacian Convolutional Neural Networks. Different from other Convolutional Neural Networks models, We put signal enhancement layer into Convolutional Neural Networks structure and Laplacian filter is used in the signal enhancement layer. We test the proposed method on four kinds of small-size image databases. The experimental results have demonstrate that the proposed algorithm is effective. The detection accuracies for different image size database are all above 95%.