摘要
视频拍摄中,常因相机抖动或拍摄对象移动造成视频模糊,给后续的信息获取及视频处理带来干扰。为了更好地利用视频帧的全局上下文信息,本文提出了一种基于金字塔池化和注意力机制的视频去模糊算法。在视频去模糊的复原网络中引入金字塔池化,利用不同尺度的池化获得更加全面的全局上下文信息;使用注意力机制加强对全局上下文信息的利用,以达到提升视频去模糊的效果。在DVD数据集上的实验结果表明,该算法能够有效地提升视频复原效果。
The video is often blurred caused by camera shake or object motions during the exposure time,which interferes with subsequent information acquisition and video processing.In order to make better use of the global context information of video frame,a video deblurring algorithm based on pyramid pooling and attention mechanism is proposed.Pyramid pool is introduced in the video deblurring reconstruction networks to obtain more comprehensive global context information by different scales pooling.Then the attention mechanism is used to enhance the use of global context information to improve the effect of video deblurring.The experimental results on DVD dataset show that the algorithm can effectively improve the video restoration effect.
作者
邹世奇
刘洪
ZOU Shiqi;LIU Hong(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)
出处
《智能计算机与应用》
2023年第12期75-79,共5页
Intelligent Computer and Applications
基金
贵州省科学技术基金(黔科合基础[2019]1063号)
贵州大学引进人才科研项目(贵大人基合同字(2017)14号)。