Enhanced motion compensated frame interpolation using object layer inference

Abstract

In motion compensated frame interpolation (MCFI), an interpolated frame is produced by placing objects according to their motion estimated between two neighbouring frames. However, multiple motion vectors associated with the background (BG) and foreground (FG) objects can pass through a single location in the interpolated frame. In such motion ambiguity (MA) regions, annoying artefacts can be induced by adopting the motion of the BG object instead of the FG object. Introduced is a novel MCFI algorithm which can accurately infer the motion of the FG object at the MA region with low complexity. Simulation results show that the proposed method achieves higher visual quality for the interpolated frame without annoying artefacts as compared with conventional MCFI algorithms.