Scholarlink

<p>The rise of ultra-high-definition (UHD) display technologies (e.g., 4K/8K) and increasingly complex streaming media environments pose significant challenges for accurately assessing Content Delivery Network (CDN) services and their associated Quality of Experience (QoE). However, current QoE assessment methods inadequately integrate both video aesthetics and technical characteristics. Moreover, existing sampling approaches for streaming video QoE evaluation suffer from inherent limitations, particularly in capturing viewing experiences affected by buffering events. To address these challenges, we propose SFQoE, a novel streaming video QoE assessment method that integrates hybrid feature fusion and adaptive buffering sampling. First, we design a novel dual-branch fusion framework that integrates a pre-trained aesthetic model (DOVER) with a Laplacian-based sharpness module. This architecture synergizes high-level aesthetic perception with low-level texture preservation, addressing a key gap in current QoE assessment methods. Second, we introduce adaptive buffered sampling, an innovative buffer-event-driven adaptive sampling approach. The sparse-buffering hybrid sampling strategy achieves computational efficiency without compromising feature completeness, overcoming conventional methods’ limitations in buffered frame capture. Extensive experimental results on four publicly available streaming video QoE datasets, namely WaterlooSQoE-I, LIVE-NFLX-II, WaterlooSQoE-III, and WaterlooSQoE-IV, demonstrate that the SFQoE method achieves highly competitive and balanced performance in both prediction accuracy and real-time processing capability.</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

Streaming video QoE assessment method combining hybrid feature fusion and adaptive buffer sampling

  • 期刊论文

  • 发表日期：

    2026年6月25日

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

摘要

The rise of ultra-high-definition (UHD) display technologies (e.g., 4K/8K) and increasingly complex streaming media environments pose significant challenges for accurately assessing Content Delivery Network (CDN) services and their associated Quality of Experience (QoE). However, current QoE assessment methods inadequately integrate both video aesthetics and technical characteristics. Moreover, existing sampling approaches for streaming video QoE evaluation suffer from inherent limitations, particularly in capturing viewing experiences affected by buffering events. To address these challenges, we propose SFQoE, a novel streaming video QoE assessment method that integrates hybrid feature fusion and adaptive buffering sampling. First, we design a novel dual-branch fusion framework that integrates a pre-trained aesthetic model (DOVER) with a Laplacian-based sharpness module. This architecture synergizes high-level aesthetic perception with low-level texture preservation, addressing a key gap in current QoE assessment methods. Second, we introduce adaptive buffered sampling, an innovative buffer-event-driven adaptive sampling approach. The sparse-buffering hybrid sampling strategy achieves computational efficiency without compromising feature completeness, overcoming conventional methods’ limitations in buffered frame capture. Extensive experimental results on four publicly available streaming video QoE datasets, namely WaterlooSQoE-I, LIVE-NFLX-II, WaterlooSQoE-III, and WaterlooSQoE-IV, demonstrate that the SFQoE method achieves highly competitive and balanced performance in both prediction accuracy and real-time processing capability.