<FONT face=Verdana>This paper proposes a blind detection technique to reveal malicious attacks on video contents by video sequence montage. The method is based on an analysis of Fourier transform of the series of motion errors. For an MPEG-2 video sequence with a particular structure of the group of pictures (GOP)

spikes exist in the Fourier transform magnitude of the error series. Multiple MPEG compression with different GOP structures results in variation of spikes in the Fourier domain

leaving clear signatures of the coding history in the video sequence. These spike artifacts are used to detect video splicing

i.e.

replacing certain sections of the video with fake contents taken from other sequences with different GOP structures. This method can also be used to detect suspicious multiple compressions and find the initial GOP structure

providing useful clues to the authentication of video contents. Experiments show effectiveness of the proposed method.