<![CDATA[Semantic segmentation is crucial step in autonomous driving, medical imaging, and scene understanding. Traditional approaches leveraging manually extracted pixel properties and probabilistic models, have achieved reasonable performance but suffer from limited generalization and the need for expert-driven feature selection. The rise of deep learning architectures has significantly improved segmentation accuracy by enabling automatic feature extraction and capturing intricate object details. However, these methods still face challenges, including the need for large datasets, extensive hyperparameter tuning, and careful loss function selection. This paper proposes a novel boundary-sensitive loss function, which combines region loss and boundary loss, to enhance both region consistency and edge delineation in segmentation tasks. Implemented within a modified SegNet framework, the approach proposed in the paper is evaluated with the semantic boundary dataset (SBD) dataset using standard segmentation metrics. Experimental results indicate improved segmentation accuracy, substantiating to proposed method.]]>
Copyrights © 2025
