跳至主導覽 跳至搜尋 跳過主要內容

Spatially adaptive and flexible deep network for blind compressed image reconstruction

研究成果: 書貢獻/報告類型會議論文篇章

摘要

JPEG is the most widely used image compression format, but low quality factors introduce blocking and ringing artifacts that degrade visual quality and hinder computer vision tasks. Existing deep learning methods often require separate models for different compression levels, whereas FBCNN (flexible blind convolutional neural network) handles multiple qualities with a single model but remains limited by ResNet (deep residual learning networks)-based blocks (ResBlocks) in capturing diverse features. To address this, we propose SA-FBCNN (a Spatially Adaptive Flexible Blind deep Convolutional Neural Network) that enhances feature extraction across regions and scales. With lower model complexity, our framework achieves superior quantitative and qualitative JPEG image reconstruction performance.

原文英語
主出版物標題International Workshop on Advanced Imaging Technology, IWAIT 2026
編輯Masayuki Nakajima, Chuan-Yu Chang, Chien-Chou Lin, Shogo Tokai, Kwang-Deok Seo, Chia-Hung Yeh, Budianto Tandianus
發行者SPIE
ISBN(電子)9798902321033
DOIs
出版狀態已發佈 - 2026 2月 27
事件International Workshop on Advanced Imaging Technology, IWAIT 2026 - Kaohsiung, 臺灣
持續時間: 2026 1月 122026 1月 14

出版系列

名字Proceedings of SPIE - The International Society for Optical Engineering
14072
ISSN(列印)0277-786X
ISSN(電子)1996-756X

會議

會議International Workshop on Advanced Imaging Technology, IWAIT 2026
國家/地區臺灣
城市Kaohsiung
期間2026/01/122026/01/14

ASJC Scopus subject areas

  • 電子、光磁材料
  • 儀器
  • 凝聚態物理學
  • 電腦科學應用
  • 應用數學
  • 電氣與電子工程

指紋

深入研究「Spatially adaptive and flexible deep network for blind compressed image reconstruction」主題。共同形成了獨特的指紋。

引用此