DRRU-Net: DCT-Coefficient-Learning RRU-Net for Detecting an Image-Splicing Forgery

In this paper, we propose a lightweight deep learning network (DRRU-Net) for image-splicing forgery detection. DRRU-Net is an architecture that combines RRU-Net the visual content of images and image acquisition artifacts, JPEG artifact module compression artifacts in discrete cosine transform (DCT) domain. The backbone model based on pre-training, such as CAT-Net, representative detection, has relatively large number parameters, resulting overfitting small dataset, which hinders generalization performance. Therefore, designed to learn characteristics according DCT domain real time without pre-training. experiments, proposed training method show parameters are smaller than detection performance better RRU-Net, various datasets can be improved.