Scholarlink

<p>The proposed methodology for video coding aims to improve efficiency by integrating advanced object detection and compression techniques. Traditional video coding methods often face challenges related to high computational complexity and inefficiency when processing large volumes of data. These methods do not effectively distinguish between significant objects and less relevant background information, leading to suboptimal compression rates. To address these limitations, the proposed approach begins by extracting video frames from a publicly available dataset, followed by object detection using an enhanced version of YOLO-v8 (En-YOLO-v8). The detection process is optimized with the Adaptive Lyrebird Optimization algorithm (AdaLyBO), which refines the model’s loss function to enhance detection accuracy. Once the objects are identified, a Generative Adversarial Network (GAN) is applied for object-aware video compression. The GAN efficiently compresses the video by focusing on objects of interest while minimizing unnecessary background details. The combination of En-YOLO-v8 for object detection and GAN-based compression significantly improves video coding efficiency compared to existing methods. The proposed method achieved an average mean Average Precision (mAP) of 30.0756%, average Mean Square Error (MSE) of 0.1659, average Multi-Scale Structural Similarity Index Measure (MS-SSIM) of 19.405 dB, average Peak Signal-to-Noise Ratio (PSNR) of 29.274 dB, Bitrate Saving of 33.89%, and Computation Time of 16.42sec.</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

Enhanced YOLO-v8 with GAN based object aware video coding for efficient compression

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Ms Arunadevi Arumugam,
  • Dr Jaisiva Selvaraj

摘要

The proposed methodology for video coding aims to improve efficiency by integrating advanced object detection and compression techniques. Traditional video coding methods often face challenges related to high computational complexity and inefficiency when processing large volumes of data. These methods do not effectively distinguish between significant objects and less relevant background information, leading to suboptimal compression rates. To address these limitations, the proposed approach begins by extracting video frames from a publicly available dataset, followed by object detection using an enhanced version of YOLO-v8 (En-YOLO-v8). The detection process is optimized with the Adaptive Lyrebird Optimization algorithm (AdaLyBO), which refines the model’s loss function to enhance detection accuracy. Once the objects are identified, a Generative Adversarial Network (GAN) is applied for object-aware video compression. The GAN efficiently compresses the video by focusing on objects of interest while minimizing unnecessary background details. The combination of En-YOLO-v8 for object detection and GAN-based compression significantly improves video coding efficiency compared to existing methods. The proposed method achieved an average mean Average Precision (mAP) of 30.0756%, average Mean Square Error (MSE) of 0.1659, average Multi-Scale Structural Similarity Index Measure (MS-SSIM) of 19.405 dB, average Peak Signal-to-Noise Ratio (PSNR) of 29.274 dB, Bitrate Saving of 33.89%, and Computation Time of 16.42sec.