Research Article
A Dual-Stream Input Faster-CNN Model for Image Forgery Detection
@INPROCEEDINGS{10.1007/978-3-031-32443-7_7, author={Lizhou Deng and Ji Peng and Wei Deng and Kang Liu and Zhonghua Cao and Wenle Wang}, title={A Dual-Stream Input Faster-CNN Model for Image Forgery Detection}, proceedings={Mobile Networks and Management. 12th EAI International Conference, MONAMI 2022, Virtual Event, October 29-31, 2022, Proceedings}, proceedings_a={MONAMI}, year={2023}, month={5}, keywords={Image Forgery Detection Faster-CNN Dual-Stream Input}, doi={10.1007/978-3-031-32443-7_7} }- Lizhou Deng
Ji Peng
Wei Deng
Kang Liu
Zhonghua Cao
Wenle Wang
Year: 2023
A Dual-Stream Input Faster-CNN Model for Image Forgery Detection
MONAMI
Springer
DOI: 10.1007/978-3-031-32443-7_7
Abstract
With the development of multimedia technology, the difficulty of image tampering has been reduced in recent years. Propagation of tampered images brings many adverse effects so that the technology of image tamper detection needs to be urgently developed. A faster-rcnn based image tamper localization recognition method with dual-flow Discrete Cosine Transform (DCT) high-frequency and low-frequency input is presented. For capturing subtle transform edges not visible in RGB domain, we extract high-frequency features from the image as an additional data stream embedding model. Our network model uses low-frequency images as the subject data to detect object consistency in different regions, further complements high-rate streams to strengthen image region consistency detection, and complements duplicate stream object tampering detection. Extensive experiments are performed on the CASIA V2.0 image dataset. These results demonstrate that faster-rcnn-w outperforms existing mainstream image tampering detection methods in different evaluation indicators.
