Foreground–background separation and video reconstruction using moment matching

We study the foreground–background separation problem for a video stream in the moment matching framework. We first show how a video stream can be represented as the steady-state output of a dynamical system. We then leverage this technique to reconstruct image sequences in a video stream. In addition, we show that foreground–background separation is a special case of this reconstruction algorithm. Finally, we highlight the computational benefits of the proposed method when compared to other methods that associate a sequence of images with a dynamical system.