Research Seminar - October 10, 2003

Video Rewrite

Abstract:

My proposed research is "video rewrite". It is a new field of research aiming to utilize the photo-realism of real video input to synthesize new video output without having three dimensional (3D) information of the scene. By finding similar frames and restructuring the original frame sequence, it is possible to generate realistic looking animations or video loops without having to construct complex 3D model and movement model. Two areas of video rewrite I am interested in are video-based animation and "video texture" generation. Video texture is a video that plays forever, which could be used as a dynamic picture.

The basic difficulty of rewriting video is how to preserve temporal and spatial coherence of the output video while the sequence and regions of input video is being restructured, either randomly or under specific control. The problem areas associated with it are segmentation, similarity measurement, transformation, representation of features, control scheme for animation and the final stitching-together.

Since this area of research is in its beginning stage, current research uses mostly off-the-shelf algorithms or implementations. More research is needed to explore this new open land. I am interested in investigation on the areas of similarity measurement, transformation techniques and animation control scheme.