摘要
视频编解码过程中,常因量化操作的引入,导致视频出现有损压缩,在后续播放中,产生较多压缩残影,且在低比特编码中更为明显。因此,针对此问题的优化,在深度学习的理论基础上,提出了一种新型的基于光流技术的时空神经网络模型(FGTSN),旨在对编码视频进行精确的后处理。实验证明,该FGTSN方法能够显著提高HEVC压缩视频的质量,其效果远超过其他视频质量增强技术。此方法能有效解决遮挡和大范围运动场景下的问题,并提升了压缩视频帧的重建效率,证明了其在实际应用中的高价值。
In the process of video encoding and decoding,the introduction of quantization operations often leads to lossy compression of the video.In subsequent playback,there are more compression residues,which are more pronounced in low bit encoding.Therefore,in order to optimize this problem,a new spatiotemporal neural network model based on optical flow technology(FGTSN)is proposed on the basis of deep learning theory,aiming to perform accurate post-processing on encoded videos.Experimental results have shown that the FGTSN method can significantly improve the quality of HEVC compressed videos,and its effectiveness far exceeds other video quality enhancement techniques.This method can effectively solve the problems of occlusion and large-scale motion scenes,and improve the reconstruction efficiency of compressed video frames,proving its high value in practical applications.
作者
孙斐然
SUN Feiran(Wujiang District Integrated Media Center,Suzhou Jiangsu 215200)
出处
《软件》
2024年第4期113-115,121,共4页
Software
关键词
深度学习理念
视频编解码技术
FGTSN方法
deep learning concept
video encoding and decoding technology
FGTSN method
