The system is based on computational rules that can be carried out on a PC. The program looks for parts of an image that are of the same color and move in the same way. These parts together form an object, for example a person, animal, car or plane.
The program has no difficulty dealing with a multi-colored object in motion because all the points move in the same direction. It uses the previous images to continuously predict what an object looks like, even if it is out of sight.
The new method unravels video images into separate objects that can be cut, altered and analyzed. Previous video manipulation systems were only able to alter colors to some extent or to cut a recording into individual sections that can be edited into a different order, for example.
Although the research is basically a fundamental science, the researchers do foresee practical applications for their findings. The computational rules can be incorporated into commercial video manipulation programs, for example, either for amateurs or professionals.