Parallel Scale-wise Attention Network for Effective Scene Text Recognition

Sajid, Usman; Zhang, Jin; Kim, Taejoon; Wang, Guanghui

doi:10.1109/ijcnn52387.2021.9534223

Cited by 8 publications

(3 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…•Big fluctuation in text appearance and scale: Some of the text may be scaled or rotated, this creates a problem in detection because there is no suitable text [5].…”

Section: Challengesmentioning

confidence: 99%

Scene Text Recognition: A Review

Mageed

Abbadi

2023

Iraqi Journal of Science

View full text Add to dashboard Cite

The problem of text recognition and its applicability as part of images captured in the wild has gained a significant attention from the computer vision community in recent years. In contrast to the recognition of printed documents, scene text recognition is a difficult problem. Contrary to recognition of printed documents, recognizing a scene text is a challenging problem. Many researches focus on the problem of recognizing text extracted from natural scene images. Significant attempts have been made to address this problem in recent past. However, many of these attempts work on utilizing availability of strong context, which naturally limits the dictionary. This paper presents a review of recent papers related to scene text recognition in the period (2013-2020). This paper helps other researchers to understand the whole system of scene text recognition instead of reading many papers in isolation.

show abstract

“…•Big fluctuation in text appearance and scale: Some of the text may be scaled or rotated, this creates a problem in detection because there is no suitable text [5].…”

Section: Challengesmentioning

confidence: 99%

Scene Text Recognition: A Review

Mageed

Abbadi

2023

Iraqi Journal of Science

View full text Add to dashboard Cite

show abstract

“…Following its success in machine translation, the attention mechanism has started to be applied for text recognition (Gao et al, 2017;Sajid et al, 2021). (Chowdhury and Vig, 2018) combined a CNN backbone with an RNN encoder-decoder for HTR.…”

Section: Related Workmentioning

confidence: 99%

“…While the CTC architecture has been successful in OCR, it still has limitations in terms of language modeling, which can be improved by using an external language model as a post-processing step. To overcome this limitation, attention mechanisms have gained popularity in text recognition tasks [11,17,26,32]. Chowdhury and Vig [11] introduced an RNN encoder-decoder model for HTR that utilized a CNN backbone.…”

Section: Related Workmentioning

confidence: 99%

Arabic Offline Handwritten Text Recognition with Transformers

Momeni

BabaAli

2022

Preprint

View full text Add to dashboard Cite

Handwriting recognition has always been one of the most important and challenging issues in pattern recognition and machine learning due to its diverse applications. In this paper, we investigate the problem of Arabic offline handwritten text recognition. Existing approaches mainly use a combination of convolutional and recurrent layers for image understanding and connectionist temporal classification for text generation. Due to the sequential nature of recurrent neural networks, these methods suffer from a lack of parallelization. In addition, since these models cannot model linguistic rules, an external language model is often used in the post-processing stage to increase accuracy. To address these problems, we consider two different architectures, namely the Transformer Transducer and the standard sequence-to-sequence Transformer, and compare them in terms of accuracy and performance. Our approach can model language dependencies and relies only on the attention mechanism, making it more parallelizable and less complex. We adopt pre-trained Transformers for both image understanding and language modeling. Evaluation results on the Arabic KHATT dataset show that our proposed method outperforms the current state-of-the-art on Arabic handwritten text recognition task.

show abstract