Development of Video Features to Detect Spatially Modified Video

Extreme growth of the Internet has made the process of data transmission, distribution and access to digital media content becomes easier. However, we can find clips of videos which one or other aspects were part of other videos without mentioning the later ones, thus infringing one's copyright. So, efforts to research and develop video copy detection has grown significantly. The existing works have focused only on trying to detect modification on content edition, removing segments of videos or inserting segments into one's videos. We still have not found any attempt to try to detect modifications in the spatial domain, for example, when the videos were cropped through the whole frames. In this research, we proposed a framework to detect modification of video on spatial domain. First, each frame was broken into blocks, Then, we extract DCT features from each block. To reduce searching computational cost, we proposed a scheme to extract keyframes. For the experiment, we have collected 1556 video clips which consist of static and dynamic scenes, and we cut up to 10% spatially on the top, bottom, left and right sides of the video clips. The experiments show that we have achieved perfect detection in 3 key frame samples and 14% spatially cut video clips.