method: CLOVA-AI v22019-02-18

Authors: Jeonghun Baek, Junyeop Lee, Sungrae Park, Moonbin Yim, Seonghyeon Kim, Hwalsuk Lee

Description: We used Thin-plate-spline (TPS) based Spatial transformer network (STN) which normalizes the input text images, ResNet based feature extractor, BiLSTM, and attention mechanism.
This model was developed based on the analysis of scene text recognition modules.
See our paper and source code.

What Is Wrong With Scene Text Recognition Model Comparisons? Dataset and Model Analysis (Accepted by ICCV 2019 oral)

Source code

method: SRIB-STRIDE (Scene Text Recognition In Device)2020-09-30

Authors: Rachit S Munjal, Arun D Prabhu, Sukumar Moharana, Nikhil Arora

Affiliation: Samsung R & D, Bangalore

Description: SRIB_STRIDE -Scene Text Recognition In Device is a very lightweight STR model designed for real-time on device usage, with selective-rotation, CNN with Global Squeeze-Excite Modules , Bi-LSTM with Projection and CTC Decoder.
Total Parameters: 0.85M supporting Latin diacritic characters as well.
Execution Time on S10 + :- 5ms ( dimension 16*128)

R. S. Munjal, A. D. Prabhu, N. Arora, S. Moharana and G. Ramena, "STRIDE: Scene Text Recognition In-Device," 2021 International Joint Conference on Neural Networks (IJCNN), 2021, pp. 1-8, doi: 10.1109/IJCNN52387.2021.9534319.

method: PIONEER2013-04-06

Authors: Jerod Weinman

Description: The Grinnell PIONEER system (Probabilistic Integrator of Noisy Evidence Enabling Reading) of Weinman et al. first segments each word image into superpixels with a mixture of position-conditional regressors. For binarization, a logistic regression classifier then chooses the mixture component(s) most likely to be text. We robustly fit second degree polynomials to the tops and bottoms of binarized characters with an EM algorithm. Word images are then normalized to a horizontal, linear orientation with a thin-plate spline. Finally, a discriminative semi-Markov model jointly segments and recognizes characters in the normalized image using steerable pyramid features, character bigrams, and a large lexicon.

Jerod J. Weinman, Zachary Butler, Dugan Knoll, and Jacqueline Feild. Toward integrated scene text reading. IEEE Trans. on Pattern Anal. Mach. Intell., 2014.

Jerod J. Weinman, Erik Learned-Miller, and Allen Hanson. A discriminative semi-Markov model for robust scene text recognition. In Proc. Intl. Conf. on Pattern Recognition, Dec 2008.

Source code

Ranking Table

Description Paper Source Code
DateMethodTotal Edit distance (case sensitive)Correctly Recognised Words (case sensitive)T.E.D. (case insensitive)C.R.W. (case insensitive)
2019-02-18CLOVA-AI v226.951595.98%23.483696.35%
2020-09-30SRIB-STRIDE (Scene Text Recognition In Device)72.792986.67%62.631487.58%
2013-04-06PIONEER479.820853.70%426.835355.71%

Ranking Graphic

Ranking Graphic