Exposing Recaptured Images with Constrained Convolutional Neural Network
Nan Zhu, Hanchen Xiang, Zhiqin Liu · 2022 7th International Conference on Signal and Image Processing (ICSIP) · 2022
Image acquisition and display technology have greatly developed over the past few years, which make it easy to recapture high-quality images from various display media, such as liquid crystal display screen or printed paper. This operation can be utilized to hide image tampering traces and eliminate image watermarking, which poses serious threat to image security. In order to solve this problem, we proposed a recaptured image detection approach based on constrained convolutional neural network. Firstly, a deep residual module is designed to extract a set of prediction error maps. This module has three identical branches, in which a $1\times 1$ convolution layer is used to learn the correlations between color channels and a following constrained convolutional layer is used to extract prediction error maps. Then we feed the output of this module into consecutive convolution modules to further learn the hierarchical representation to make final decision. Ablation experiments verify the effectiveness of our proposed deep residual module. Comparison experiments on three public databases demonstrate the superiority of our proposed method on detection accuracy.