基于时空信息感知的数字电影插帧算法研究OA

Deep learning-based video frame interpolation algorithms aim to enhance the frame rate of digital movies by gen-erating intermediate frames,thereby improving visual smoothness.However,models trained on real-world datasets still face significant challenges in generalizing to the unique artistic content found in digital movies.To address this issue,this paper first constructs a dataset Movie-VFI tailored for digital movie frame interpolation,followed by proposing a spatio-tem-poral information perception-based digital movie frame interpolation algorithm.This algorithm leverages a spatio-temporal enhanced encoding-decoding network,which not only accurately fits inter-frame motion,but also comprehensively cap-tures the visual spatial structure of the movie content.Experimental results demonstrate that the proposed algorithm signifi-cantly improves the interpolation quality on movie datasets,and the generated intermediate frames effectively preserve the visual consistency of the frame sequence.