基于光流估计的轻量级视频插帧算法

Lightweight video frame interpolation algorithm based on optical flow estimation

为了提升流媒体视频的帧率,解决视频插帧算法模型复杂的问题,提出了一种基于光流估计的轻量级视频插帧算法,简称SKFEVI算法。该算法只需进行一次光流估计,即可计算出两个视频帧的中间流信息,提升了中间流的估计速度,并在此基础上引入注意力机制进行特征融合,让模型对生成不同尺度特征的信息进行整合,增强了特征表达能力,提升了信息处理效率。实验结果表明,该算法的PSRN值提升至35.56 dB,SSIM值提升至0.978,合成的中间帧质量更好,流媒体视频播放更加流畅。

In order to improve the frame rate of streaming video and solve the problem of complex video interpolation algorithm model,a lightweight video frame interpolation algorithm based on optical flow estimation,referred to as SKFEVI,was proposed.The algorithm only needs to perform optical flow estimation once to calculate the intermediate stream information of two video frames directly,which improves the estimation speed of the intermediate stream,and on this basis,the attention(SKNet)mechanism was introduced for feature fusion,which enabled the model to integrate the feature information generated at different scales,enhanced the ability of feature expression and improves the efficiency of information processing.The experimental results show that the PSRN value and the SSIM value of the algorithm is increased to 35.56 dB and 0.978 respectively.The quality of the synthesized intermediate frame is better,which makes the streaming video play more smoothly.