Dual-Topological Learning for Structural Integrity Preservation in Image Segmentation

整理番号	2026a035
種別	一般研究-短期共同研究
研究計画題目	Dual-Topological Learning for Structural Integrity Preservation in Image Segmentation
研究代表者	KANG MEIYAN（Ajou University, Korea・Ph.D Student）
研究分野のキーワード	Topological Data Analysis (TDA), Persistent Homology, Dual-Topological Learning, Shape-aware Segmentation, Structural Integrity, Deep Learning
目的と期待される成果	The objective of this research is to develop a dual-objective learning framework that ensures both topological consistency and geometric shape fidelity in image segmentation. Standard pixel-wise loss functions often fail to preserve thin structures, leading to broken connectivity in complex data. By integrating Topological Loss (based on Persistent Homology) with Shape-aware constraints, this research aims to prevent structural defects at the source. The expected result is a robust segmentation pipeline capable of extracting seamless objects from intricate medical and engineering imagery, significantly improving the reliability of automated structural analysis.
組織委員(研究集会) 参加者(短期共同利用)	Shizuo Kaji（Kyoto University・Professor） Liu ran（Hebei Normal University，China・Ph.D Student）