Scholarlink

<p>Under the advancements in the video industry and computer technology, the scale of video data is becoming increasingly large. High-resolution videos are often constrained by bandwidth and storage conditions during storage and transmission, leading to the use of low-resolution videos. To effectively enhance video quality, this study proposes a processing framework based on AI-coded artifact restoration and video super-resolution collaboration, aimed at solving the problem of multiple degraded video quality enhancement. By downsampling high-resolution videos to generate low-resolution videos and encoding them using the HM encoder, the transmission process of low-resolution encoded videos can be simulated. A single-stage processing network is proposed in the study, which can simultaneously handle downsampling and encoding artifacts, directly converting low-resolution encoded videos into high-resolution, high-quality videos. Experimental results show that under the condition of QP = 32 on the HEVC test set, the proposed method achieves a peak signal-to-noise ratio (PSNR) of 28.62&#xa0;dB and a structural similarity index (SSIM) of 0.79, both outperforming traditional two-stage methods. On the Vid4 test set with QP = 32, the proposed method achieves a PSNR of 23.92&#xa0;dB and an SSIM of 0.64, approximately 0.33&#xa0;dB higher than other methods, demonstrating significant performance improvement on both the HEVC and Vid4 test sets. Ablation experiments show that removing the feature reuse structure results in a PSNR decrease of 0.48&#xa0;dB, and removing the intermediate transition label leads to a PSNR decrease of 0.27&#xa0;dB, verifying the significant contribution of the feature reuse structure and the intermediate transition label to the network’s performance. The proposed processing framework based on AI-coded artifact restoration and video super-resolution collaboration effectively solves the problem of quality enhancement for low-resolution videos affected by both downsampling and encoding artifacts, providing a new approach for video quality improvement.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A method for solving the multiple degradation video quality enhancement problem: a processing framework for AI-based coding damage repair in concert with video super-resolution

  • 期刊论文

  • 发表日期：

    2025年1月7日

查看全文
  • Maojin Sun

摘要

Under the advancements in the video industry and computer technology, the scale of video data is becoming increasingly large. High-resolution videos are often constrained by bandwidth and storage conditions during storage and transmission, leading to the use of low-resolution videos. To effectively enhance video quality, this study proposes a processing framework based on AI-coded artifact restoration and video super-resolution collaboration, aimed at solving the problem of multiple degraded video quality enhancement. By downsampling high-resolution videos to generate low-resolution videos and encoding them using the HM encoder, the transmission process of low-resolution encoded videos can be simulated. A single-stage processing network is proposed in the study, which can simultaneously handle downsampling and encoding artifacts, directly converting low-resolution encoded videos into high-resolution, high-quality videos. Experimental results show that under the condition of QP = 32 on the HEVC test set, the proposed method achieves a peak signal-to-noise ratio (PSNR) of 28.62 dB and a structural similarity index (SSIM) of 0.79, both outperforming traditional two-stage methods. On the Vid4 test set with QP = 32, the proposed method achieves a PSNR of 23.92 dB and an SSIM of 0.64, approximately 0.33 dB higher than other methods, demonstrating significant performance improvement on both the HEVC and Vid4 test sets. Ablation experiments show that removing the feature reuse structure results in a PSNR decrease of 0.48 dB, and removing the intermediate transition label leads to a PSNR decrease of 0.27 dB, verifying the significant contribution of the feature reuse structure and the intermediate transition label to the network’s performance. The proposed processing framework based on AI-coded artifact restoration and video super-resolution collaboration effectively solves the problem of quality enhancement for low-resolution videos affected by both downsampling and encoding artifacts, providing a new approach for video quality improvement.