Video coding post filtering method based on convolutional neural network

Abstract

The invention provides a video coding post filtering method based on a convolutional neural network. The method comprises a convolutional neural network model training step and a filtering step, the training step comprises: setting a quantization parameter of video compression to 20 to 51, and performing coding compression on the original video to obtain a compressed video; performing frame extraction on all videos to obtain a plurality of frame pairs of compressed video frames and original video frames; dividing the extracted frame pairs into a plurality of groups according to the frame types and different quantization parameters; establishing a convolutional neural network framework and initializing the network parameters, and training the neural network by using the divided groups to obtain a plurality of convolutional neural network models corresponding to different quantization parameters and frame types. The filtering step comprises: executing the foregoing coding compression and frame extraction on the to-be-processed original video to obtain to-be-processed frame pairs, and selecting the corresponding convolutional neural network models according to the quantization parameters and the frame types of the to-be-processed frame pairs to perform filtering processing.

Description

technical field [0001] The invention relates to the fields of computer vision and video coding, in particular to a convolutional neural network-based post-filtering method for video coding. Background technique [0002] With the development of science and technology and the abundance of various video display devices, video has increasingly become an indispensable part of people's lives and plays a very important role in various fields. The past few decades have witnessed the tremendous development of video resolution and display device screens, and ultra-high-resolution video will generate a huge amount of data, which will impose a great burden on network bandwidth. Therefore, high-efficiency video encoding and transmission technologies are required to ensure the user experience of watching videos, and at the same time reduce the amount of video data as much as possible to reduce the burden on network bandwidth. For this reason, researchers have been studying efficient vide...

AI Technical Summary

Problems solved by technology

Although some algorithms have been adopted in the current video coding standards to reduce the block effect and improve the subjective quality, the effect is not ideal. The processed video still has obvious block effect and edge blur, and the loss of details is still relatively serious.

Images