Gated fusion network for SAO filter and inter frame prediction in Versatile Video Coding

In order to achieve higher coding efficiency, the Versatile Video Coding (VVC) standard includes several new components at expense of an increase in decoder computational complexity. These technologies often create ringing and contouring effects on reconstructed frames a low bit rate introduce blurring distortion. To smooth those visual artifacts, H.266/VVC framework supports four post-processing filter operations. The state-of-the-art CNN-based in-loop filters prefer deploy multiple networks for various quantization parameters frame resolutions, which increases training resources subsequently becomes overhead reconstruction. This paper presents single deep-learning-based model sample adaptive off-set (SAO) non-linear filtering operation side, uses feature correlation among adjacent frames, substantiates merits intra–inter quality enhancement. We introduced variable size dual multi-scale convolutional neural network (D-MSCNN) attenuate compression artifact incorporated strided deconvolution restore high-frequency details distorted frame. Our follows sequential across all QP values updates weights. Using data augmentation, weight fusion, residual learning, we demonstrated that our could be trained effectively by transferring convolution prior indices produce dense output map. Objective measurements demonstrate proposed method outperforms baseline VVC PSNR, MS-SSIM, VMAF metrics achieves average 5.16% saving different test sequence categories.