A Concatenated Model for Video Frame Interpolation

Abstract

A concatenated model for video frame interpolation is presented in this paper. The algorithm involves warping a high resolution reference image in a way that preserves much of the high resolution detail in the sequence. The concatenated model incorporates adaptive bidirectional warping and has the flexibility to incorporate a variety of frame-based interpolation methods within the system. The concatenated system is evaluated using several constituent interpolation methods in conjunction with the adaptive warping. It is shown that high subjective quality and PSNRs can be obtained relative to the benchmark enlargement methods considered.