Abstract. The paper presents a naturalness preserving transform (NPT) based

collusion and compression resistant watermarking technique for video. An image

that is statistically similar to a video frame is chosen as the watermark and

this image is embedded independently in consecutive frames of the video. To

enhance the resistance to inter frame collusion based attacks, a non-overlapping

block matching is used to determine the region for placing the watermark in

consecutive frames. Only a trace of the watermark image is embedded which

enhances the robustness of the watermark to different attacks. When a frame

and the image become substantially different, another image is chosen as the

watermark. The size of the watermark determines the quality of the watermarked

video frames. Watermark extraction is blind and requires only the region

where the watermark was originally placed. The reconstruction process is

iterative and bestows immunity the watermark against noise and lossy compression.

Analysis indicates that the watermark is sufficiently immune to second

order inter-frame statistical attacks and is quite robust to image level compression.

Experimental results confirm these theoretical findings and demonstrate

the resistance of the technique to temporal frame averaging, additive noise and

JPEG based compression. However, the technique is limited by the fact that the

original video sequence (frames) is required for reconstruction based recovery

of the watermark from the watermarked video sequence.