An Artifact Removal Algorithm Based on CBAM-DnGAN for VVC

By studying the removal direction of video compression artifacts, a loop filtering algorithm based on spatial channel attention GAN is proposed, which effectively extracts low-frequency and high-frequency features, and integrates the attention mechanism, which can establish a global dependency relationship, expand the receptive field of the image, and obtain more context information. In addition, this paper studies the shallow network layer of VGG-22, focuses on the extracted low-frequency features, and defines the joint perception loss together with the network layer used for high-frequency feature acquisition such as VGG-54, so that the reconstructed image can not only have the characteristics of smooth details, but also reduce different types of compression artifacts, so that the subjective image has better visual effects, and at the same time, it can also achieve better index results in objective evaluation. Experimental results show that compared with VTM15.0 reference software, the BD-Rate of the proposed algorithm for video sequences with different resolutions under Y, U and V components is reduced by 1.72%, 7.09% and 8.51%, respectively.