This well-written paper discusses the authors’ efforts at attacking a very difficult problem--how to identify and then modify objects in video. The modifications can consist simply of editing the objects or of modifying the path of objects as time goes on in the video. According to the authors, it is not really possible to accomplish this completely automatically. On a very basic level, the creator or video editor at least has to manually specify what modifications he or she desires.

The authors describe their proposal. This is a generalized process that allows for objects to be identified and then either created or modified. (It also allows objects to be represented using their representation in MPEG 4.) The authors’ techniques involve establishing a scenario of what a creator might want to accomplish. Then, scenes are defined as video object planes (VOPs) on top of fixed backgrounds. Next, the authors define what they call a warping model in two dimensions, which allows for a simulation of how objects can move in three dimensions. Then, they define a method that allows for the automated definition of a zoom factor for the VOPs. While the process described in the paper has some limitations, such as difficulty in rotating objects around an axis in some situations, it is demonstrated to be successful in many situations.