Scholarlink

<p>High Efficiency Video Coding (HEVC), created through joint industry‑academic efforts, offers significant improvement in video compression efficiency but requires substantially greater computational resources. This is because of the use of recursive quad-tree partitioning structure. This quad-tree approach remains central in recent standards such as Versatile Video Coding (VVC). In this paper, we introduce the Quantum Tree Partition Network (QTPNet), a hybrid quantum-classical network designed to predict quad-tree partitioning decisions in HEVC intra-prediction mode focused on a Quantization Parameter (QP) value of 32. Unlike classical Convolutional Neural Networks (CNNs), QTPNet integrates simulated quantum circuits to produce nonlinear feature representations. These quantum-circuit-simulated features allow more effective modeling of spatial dependencies within quad-tree blocks, especially improving prediction accuracy for smaller quad-tree blocks with limited spatial context, where traditional CNNs typically underperform. Experimental evaluations, performed by simulating various quantum-circuit configurations on classical hardware, demonstrate that QTPNet achieves an average encoding time reduction of 69.54% and a Bjøntegaard Delta Bit-Rate (BD-BR) improvement of -0.99% compared to the HEVC reference software (HM-16.5), without significant loss in Bjøntegaard Delta Peak Signal-to-Noise Ratio (BD-PSNR).</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

Hybrid quantum CNN applied to quad-tree decision on HEVC

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Iris Linck,
  • Arthur Tórgo Gómez,
  • Gita Alaghband

摘要

High Efficiency Video Coding (HEVC), created through joint industry‑academic efforts, offers significant improvement in video compression efficiency but requires substantially greater computational resources. This is because of the use of recursive quad-tree partitioning structure. This quad-tree approach remains central in recent standards such as Versatile Video Coding (VVC). In this paper, we introduce the Quantum Tree Partition Network (QTPNet), a hybrid quantum-classical network designed to predict quad-tree partitioning decisions in HEVC intra-prediction mode focused on a Quantization Parameter (QP) value of 32. Unlike classical Convolutional Neural Networks (CNNs), QTPNet integrates simulated quantum circuits to produce nonlinear feature representations. These quantum-circuit-simulated features allow more effective modeling of spatial dependencies within quad-tree blocks, especially improving prediction accuracy for smaller quad-tree blocks with limited spatial context, where traditional CNNs typically underperform. Experimental evaluations, performed by simulating various quantum-circuit configurations on classical hardware, demonstrate that QTPNet achieves an average encoding time reduction of 69.54% and a Bjøntegaard Delta Bit-Rate (BD-BR) improvement of -0.99% compared to the HEVC reference software (HM-16.5), without significant loss in Bjøntegaard Delta Peak Signal-to-Noise Ratio (BD-PSNR).