Using background memory for efficient video coding

In the majority of video conference applications, the camera position is fixed. The composition of the background of the scene is usually stationary. When the subject moves in the scene, some parts of the background may be occluded and some other parts of the background may be uncovered. Such uncovered background parts usually has to be encoded using intra mode coding. Furthermore, if a pair of cameras is used, for example, one for the speaker and the other for showing a document, the entire scene change frame has to be intra frame encoded whenever a scene cut happens. However, either the uncovered background or scene change frame may have been transmitted in the previous frames. We propose a background memory update scheme which addresses the above problem in a video codec so that a higher coding efficiency under such a scenario can be achieved. The experimental results show that the proposed scheme can achieve up to a 0.7 dB PSNR improvement in the rate-distortion plane.