Robust Scene Text Detection via Learnable Scene Transformations

Yuheng Cao, Mengjie Zhou, Jie Chen
Proceedings of The 14th Asian Conference on Machine Learning, PMLR 189:137-152, 2023.

Abstract

Scene text detection based on deep neural networks has been extensively studied in the last few years. However, the task of detecting texts in complex scenes such as bad weather and image distortions has not received sufficient attentions in existing works, which is crucial for real-world applications such as text translation, autonomous driving, etc. In this paper, we propose a novel strategy to automatically search for the effective scene transformation polices to augment images in the training phase. In addition, we build a new dataset, Robust-Text, to evaluate the robustness of text detection methods in real complex scenes. Experiments conducted on the ICDAR2015, MSRA-TD500 and Robust-Text datasets demonstrate that our method can effectively improve the robustness of text detectors in complex scenes.

Cite this Paper

BibTeX
@InProceedings{pmlr-v189-cao23a, title = {Robust Scene Text Detection via Learnable Scene Transformations}, author = {Cao, Yuheng and Zhou, Mengjie and Chen, Jie}, booktitle = {Proceedings of The 14th Asian Conference on Machine Learning}, pages = {137--152}, year = {2023}, editor = {Khan, Emtiyaz and Gonen, Mehmet}, volume = {189}, series = {Proceedings of Machine Learning Research}, month = {12--14 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v189/cao23a/cao23a.pdf}, url = {https://proceedings.mlr.press/v189/cao23a.html}, abstract = {Scene text detection based on deep neural networks has been extensively studied in the last few years. However, the task of detecting texts in complex scenes such as bad weather and image distortions has not received sufficient attentions in existing works, which is crucial for real-world applications such as text translation, autonomous driving, etc. In this paper, we propose a novel strategy to automatically search for the effective scene transformation polices to augment images in the training phase. In addition, we build a new dataset, Robust-Text, to evaluate the robustness of text detection methods in real complex scenes. Experiments conducted on the ICDAR2015, MSRA-TD500 and Robust-Text datasets demonstrate that our method can effectively improve the robustness of text detectors in complex scenes.} }
Endnote
%0 Conference Paper %T Robust Scene Text Detection via Learnable Scene Transformations %A Yuheng Cao %A Mengjie Zhou %A Jie Chen %B Proceedings of The 14th Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Emtiyaz Khan %E Mehmet Gonen %F pmlr-v189-cao23a %I PMLR %P 137--152 %U https://proceedings.mlr.press/v189/cao23a.html %V 189 %X Scene text detection based on deep neural networks has been extensively studied in the last few years. However, the task of detecting texts in complex scenes such as bad weather and image distortions has not received sufficient attentions in existing works, which is crucial for real-world applications such as text translation, autonomous driving, etc. In this paper, we propose a novel strategy to automatically search for the effective scene transformation polices to augment images in the training phase. In addition, we build a new dataset, Robust-Text, to evaluate the robustness of text detection methods in real complex scenes. Experiments conducted on the ICDAR2015, MSRA-TD500 and Robust-Text datasets demonstrate that our method can effectively improve the robustness of text detectors in complex scenes.
APA
Cao, Y., Zhou, M. & Chen, J.. (2023). Robust Scene Text Detection via Learnable Scene Transformations. Proceedings of The 14th Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 189:137-152 Available from https://proceedings.mlr.press/v189/cao23a.html.

Related Material