Text Recognition in UAV Aerial Images

Shu Wang, Dianwei Wang, Pengfei Han, Xincheng Ren, Zhijie Xu

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Citations (Scopus)


Text recognition in unmanned aerial vehicle (UAV) aerial images is an important branch in the field of machine intelligence, which can provide important discriminative information for subsequent applications. At this stage, text recognition methods have made breakthrough progress, but the recognition of distorted and slanted text is still a challenge. In this case, we construct a text recognition network model with correction module, and propose a new type of UAV aerial image text recognition method. Specifically, the model mainly includes two parts: rectification network and recognition network. The rectification network can be optimized without manual annotation, and it can regularize various distorted and inclined UAV image texts. The recognition network introduces the attention mechanism and improves the decoder to perform bidirectional recognition of the rectified UAV image text. In addition, we verify the effectiveness of the rectification network through a large number of experiments, and prove that the model composed of the rectification network and the recognition network can achieve the optimal recognition performance.

Original languageEnglish
Title of host publicationAIPR 2021 - 2021 4th International Conference on Artificial Intelligence and Pattern Recognition
PublisherAssociation for Computing Machinery (ACM)
Number of pages7
ISBN (Electronic)9781450384087
Publication statusPublished - 24 Sep 2021
Event4th International Conference on Artificial Intelligence and Pattern Recognition - Virtual, Online, China
Duration: 24 Sep 202126 Sep 2021
Conference number: 4


Conference4th International Conference on Artificial Intelligence and Pattern Recognition
Abbreviated titleAIPR 2021
CityVirtual, Online
Internet address


Dive into the research topics of 'Text Recognition in UAV Aerial Images'. Together they form a unique fingerprint.

Cite this