Experiment on video watermark detection system using degraded original images

Watermarks often have to be detected from videos that have experienced such severe image processing as combinations of re-encoding and resizing in real applications. To improve watermark detection performance, this paper proposes a new method consisting of the following four steps: (a) estimate the image processing the watermarked images experienced, (b) apply the estimated image processing to the original images, (c) subtract the degraded original images from the watermarked images, and (d) detect watermarks from the difference images. Under the redistributing threat model, experimental evaluations show that the proposed method yields a detection ratio 34.4-percentage points higher than that yielded by the conventional method M1, which does not use original images.