Seeing Text Clearly in the Dark: Unifying Localization and Enhancement via Deep Morphological Synergy
IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026 · 1-22
DOI: 10.1109/tpami.2026.3753279 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. Verfasst von Chengpei Xu und 6 weiteren Autor:innen.
Abstract (Originalsprache)
Low-light text localization, an interdisciplinary field, has gained significant attention recently. The goal is to provide enhanced text locational information beneficial for both machine and human vision. Existing one-stage and two-stage methods either rely on single detectors specially designed for low-light conditions or on a simple cascade of detectors and enhancers, both of which exhibit limitations in effectively addressing the joint task of enhancement and text localization simultaneously. Relying solely on a single set of dual operations in deep morphology, we link the joint task via a bidirectional morphological sync bridge that ensures the simultaneous flow of two directions: one preserving the topological connectivity of text and the other improving its exposure level under low-light conditions. We then introduce a novel bi-level synergistic learning strategy, where text localization serves as the upper-level task and enhancement as the lower-level task. This strategy leverages bidirectional connectivity in the deep morphological sync bridge as a learnable constraint and alternately updates coupled gradients with different update frequencies to optimize the joint task. Extensive experiments on the low-light text localization dataset LATeD demonstrate that the proposed method outperforms state-of-the-art techniques in both tasks. The code will be released.
Zitieren
APA
Xu, C., Han, G., Zheng, F., Ma, L., Wang, W., Luo, X., & Lan, R. (2026). Seeing Text Clearly in the Dark: Unifying Localization and Enhancement via Deep Morphological Synergy. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1-22. https://doi.org/10.1109/tpami.2026.3753279
Vancouver
Xu C, Han G, Zheng F, Ma L, Wang W, Luo X, et al. Seeing Text Clearly in the Dark: Unifying Localization and Enhancement via Deep Morphological Synergy. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2026:1-22. doi:10.1109/tpami.2026.3753279
BibTeX
@article{Xu2026,
author = {Xu, Chengpei and Han, Guanchao and Zheng, Fuchen and Ma, Long and Wang, Weimin and Luo, Xiaonan and Lan, Rushi},
title = {Seeing Text Clearly in the Dark: Unifying Localization and Enhancement via Deep Morphological Synergy},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
year = {2026},
pages = {1--22},
doi = {10.1109/tpami.2026.3753279}
}
EUVIMED