STDFormer: transformer-based network for arbitrary-shaped text detection in natural scenes

Jiale Su; Chongyang Zhang

doi:10.1117/12.2659991

3 February 2023 STDFormer: transformer-based network for arbitrary-shaped text detection in natural scenes

Jiale Su, Chongyang Zhang

Proceedings Volume 12511, Third International Conference on Computer Vision and Data Mining (ICCVDM 2022); 125112T (2023) https://doi.org/10.1117/12.2659991
Event: Third International Conference on Computer Vision and Data Mining (ICCVDM 2022), 2022, Hulun Buir, China

Abstract

Natural scene text detection refers to locating and representing the text in natural scene images. The existing methods of natural scene text detection are based on convolutional neural network (CNN), but it is vulnerable to useless background noise in the process of extracting the features of curved text instances because the convolution kernel of CNN is fixed in size and rectangular in shape. In order to solve this problem, this paper proposes a novel Transformer-based Feature Fusion Module (TFFM) by integrating the transformer structure into feature pyramid network to reduce the influence of background noise in the process of feature fusion. On this basis, combined with the backbone and detection head of transformer structure, a network of natural scene text detection with full transformer structure is constructed. The method proposed in this paper achieves the state-of-the-art result on CTW1500 and Total-Text datasets, and the Transformer-based Feature Fusion Module (TFFM) proposed in this paper can be easily applied to other target detection frameworks in theory.

Citation Download Citation

Jiale Su and Chongyang Zhang "STDFormer: transformer-based network for arbitrary-shaped text detection in natural scenes", Proc. SPIE 12511, Third International Conference on Computer Vision and Data Mining (ICCVDM 2022), 125112T (3 February 2023); https://doi.org/10.1117/12.2659991

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available