Scholarlink

<p>With the rapid advancement of streaming media technology, adaptive video streaming quality of experience (QoE) has become a key factor in optimizing adaptive bitrate and compression algorithms. However, distortions such as video compression artifacts and rebuffering frequently occur during the generation and transmission of adaptive video streaming, which poses a significant challenge to accurately assessing QoE. To address this challenge, we propose a novel multi-layer feature perception method (MLFP). The MLFP framework consists of three layers, i.e., frame-level perception layer, segment-level perception layer, and global perception layer. In the frame-level perception layer, the mask attention mechanism and ResNet50 capture users’ nuanced sensations of compression and transmission distortions. To effectively characterize the impact of rebuffering on the visual experience, an adaptive streaming rebuffering perception discriminator is designed. The segment-level perception layer identifies four key quality of service (QoS) features and employs the one-dimensional group convolution and GRU network to capture the intrinsic connections between these features. The global perception layer uses a streamlined neural network to handle global statistical QoS features, providing comprehensive insights into overall QoE. The perceptual scores from the three layers are fused to generate a final QoE score. The results of experiments on four public adaptive video streaming datasets, namely WaterlooSQoE-I, LIVE-NFLX-II, WaterlooSQoE-III, and WaterlooSQoE-IV, demonstrate that the proposed method can achieve effective QoE assessment in real distortion and outperforms other partially recent QoE methods.</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

Adaptive video streaming quality of experience assessment based on multi-layer feature perception

  • 期刊论文

  • 发表日期：

    2025年2月24日

查看全文
  • Jing Huang,
  • Guangqian Kong,
  • Xun Duan,
  • Huiyun Long

摘要

With the rapid advancement of streaming media technology, adaptive video streaming quality of experience (QoE) has become a key factor in optimizing adaptive bitrate and compression algorithms. However, distortions such as video compression artifacts and rebuffering frequently occur during the generation and transmission of adaptive video streaming, which poses a significant challenge to accurately assessing QoE. To address this challenge, we propose a novel multi-layer feature perception method (MLFP). The MLFP framework consists of three layers, i.e., frame-level perception layer, segment-level perception layer, and global perception layer. In the frame-level perception layer, the mask attention mechanism and ResNet50 capture users’ nuanced sensations of compression and transmission distortions. To effectively characterize the impact of rebuffering on the visual experience, an adaptive streaming rebuffering perception discriminator is designed. The segment-level perception layer identifies four key quality of service (QoS) features and employs the one-dimensional group convolution and GRU network to capture the intrinsic connections between these features. The global perception layer uses a streamlined neural network to handle global statistical QoS features, providing comprehensive insights into overall QoE. The perceptual scores from the three layers are fused to generate a final QoE score. The results of experiments on four public adaptive video streaming datasets, namely WaterlooSQoE-I, LIVE-NFLX-II, WaterlooSQoE-III, and WaterlooSQoE-IV, demonstrate that the proposed method can achieve effective QoE assessment in real distortion and outperforms other partially recent QoE methods.