In?Loop Filtering Methods For Short Videos

With the development of communication and computer technology,the explosive development of traditional media industries such as digital TV and video conferencing has brought huge challenges to data storage and network transmission.Digital video compression coding has played a huge role in it.For example,H.265 / HEVC(High Efficiency Video Coding),which is widely used in high-efficiency video coding,and the newly released H.266 / VVC(Versatile Video Coding)video framework.At the same time,with the advent of artificial intelligence and the 5G era,people's living methods have undergone tremendous changes,and fragmentation time has increased,so that new multimedia applications including short videos have gradually enriched people's daily life.It is suitable for terminal playback,so it brings new challenges to video encoding.Specifically,short video has low resolution and can be processed from the frame level,but the current encoder is not optimized based on this point.In addition,short videos are mostly played on the mobile side,which has certain requirements on the complexity of the codec side.Furthermore,in recent years,with the development of neural networks and artificial intelligence technologies,in the multimedia field,artificial intelligence technologies such as deep learning represented by convolutional neural networks have achieved remarkable results.Therefore,this paper focuses on the current challenges faced by short video,combined with short video features,on the basis of HEVC and VVC video coding framework,starting from the loop filter module,using deep learning technology to improve the performance of loop filtering in short video coding.The specific work focuses on two aspects.First,this paper combines a neural network technology to design a lightweight filtering model based on the Inception unit.The number of model layers balances the relationship between network complexity and output performance,and is finally embedded in HEVC to form a complete video frame.This model can improve the objective quality of short video sequence encoding and reduce the bit rate.Second,based on the latest VVC coding framework,this paper designs a filter network model based on dense connection blocks.Considering the complexity of the network,combined with the technical analysis in the filter module,the designed filter model replaces the SAO(Sample Adaptive Offset)And ALF(Adaptive Loop Filter)placed after the DF of VVC video coding technology.Compared with the standard software of VVC,the method in this paper also achieves the efficient encoding of short video.