Home // IARIA Congress 2025, The 2025 IARIA Annual Congress on Frontiers in Science, Technology, Services, and Applications // View article

Deep Learning Approach for Shadow Removal Using Semantic Segmentation and Attention Mechanism

Authors:
Po-Chin Chang
Shi-Jinn Horng

Keywords: Shadow removal; Area attention; Shadow region restoration; SSIM; RMSE.

Abstract:
This paper presents a novel architecture for shadow removal that leverages semantic segmentation to divide the image into distinct regions: shadow areas, foreground areas, and shadow boundaries. To capture the intricate interactions among these regions, the model incorporates a self-attention mechanism. To tackle the persistent issue of shadow boundary residues found in existing models, this approach introduces a shadow feature fusion mechanism. This mechanism employs area attention to accurately blend features across different regions, enhancing the natural transition at shadow edges and improving shadow region restoration quality. Experimental results on public datasets validate the model’s effectiveness in shadow recovery and detail preservation, as evidenced by metrics such as Structural Similarity Index Measure (SSIM) and Root Mean Square Error (RMSE). Additionally, the model demonstrates strong generalization across various test settings, highlighting its practical applicability for shadow removal tasks.

Pages: 1 to 7

Copyright: Copyright (c) IARIA, 2025

Publication date: July 6, 2025

Published in: conference

ISBN: 978-1-68558-284-5

Location: Venice, Italy

Dates: from July 6, 2025 to July 10, 2025