Motion vector-based film mode detection for frame rate up-conversion

Abstract

This paper proposes a novel film mode detection method to improve the accuracy of frame rate up-conversion (FRUC). The existing methods need to calculate the field differences which are not necessary for FRUC. Furthermore, they prone to miss a non-film mode because they focus on detecting representative film modes, e.g., 3:2/2:2 pull-down. To avoid calculating field differences in film mode detection, the proposed method utilizes motion vectors which are already obtained from the FRUC. Furthermore, to improve the accuracy of detecting a non-film mode, the proposed method defines several features which use the variation of motion vectors for each frame. Then, it compares with the patterns of film modes. The proposed method can detect a non-film mode with 89.15% detection accuracy and increases the accuracy up to 10.93% for film modes compared to benchmark methods.