Citation

Lim, Choon Chen and Loh, Yuen Peng and Wong, Lai Kuan and Huang, Ching Chun (2026) LECY-Net: Joint Low-Light Image Enhancement and Classification Y-Net. IEEE Access, 14. pp. 79685-79702. ISSN 2169-3536

Text IEEE Xplore Full-Text PDF_6.pdf - Published Version Restricted to Repository staff only Download (4MB)

Abstract

Research on enhancing images captured in ill-light environments has become common nowadays, to improve the quality and illumination of low-light images (LLIs). Nonetheless, most of the approaches emphasize solely on LLI enhancement, ignoring the influence of the enhanced images on high-level computer vision tasks. Thereby, we tackle this issue by proposing a Y-Net architecture, namely Low-light image Enhancement and Classification Y-Net (LECY-Net), to execute the LLI enhancement and classification tasks concurrently while also serving to explore the potential correlation between these interconnected tasks. The integrated framework offers a shared backbone that extracts the common features for the subsequent enhancement and classification branches, promoting feature sharing while efficiently optimizing both tasks. Meanwhile, the proposed network is further optimized by integrating a supplementary convolutional block (SCB) to preserve image details and suppress noises. Several benchmark datasets, including LOL, Microsoft COCO, and ExDark datasets, are applied to evaluate the robustness of our model. The findings highlight that our model offers an effective solution for perceptual quality enhancement in comparison to other standalone LLI enhancement methods. When compared to the only joint enhancement and classification network that also generates simultaneous outputs, our strategy presents its clear advantage in both enhancement and classification fields, particularly on the COCO dataset. Additionally, ablation studies justified that the integrated network is advantageous for establishing a correlation between visual appearance enhancements and object classifications, while substantially boosting their performances.

Item Type:	Article
Uncontrolled Keywords:	Deep learning
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	08 Jun 2026 00:54
Last Modified:	08 Jun 2026 00:54
URII:	http://shdl.mmu.edu.my/id/eprint/16097

Downloads

Edit (login required)