Learning Rich Features for Image Manipulation Detection

• DOI: 10.48550/arxiv.1805.04953
• Publication Date: 2018-01-01
• Authors: Peng Zhou; Xintong Han; Vlad I. Morariu; Larry S. Davis
• Keywords: Robustness; RGB color model; Fuse (electrical)

Image manipulation detection is different from traditional semantic object because it pays more attention to tampering artifacts than image content, which suggests that richer features need be learned. We propose a two-stream Faster R-CNN network and train endto- end detect the tampered regions given manipulated image. One of two streams an RGB stream whose purpose extract input find like strong contrast difference, unnatural boundaries, so on. The other noise leverages extracted steganalysis rich model filter layer discover inconsistency between authentic regions. then fuse through bilinear pooling further incorporate spatial co-occurrence these modalities. Experiments on four standard datasets demonstrate our framework outperforms each individual stream, also achieves state-of-the-art performance compared alternative methods with robustness resizing compression.