Offline Handwritten Text Recognition Based on CTC-Attention

Yangyang Ma; Bing Xiao

doi:10.3788/LOP202158.1210007

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 12 >Page 1210007 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 12, 1210007 (2021)

Offline Handwritten Text Recognition Based on CTC-Attention

Yangyang Ma and Bing Xiao^*

Author Affiliations

College of Computer Science, Shaanxi Normal University, Shaanxi, Xi’an, 710062 China

show less

DOI: 10.3788/LOP202158.1210007 Cite this Article Set citation alerts

Yangyang Ma, Bing Xiao. Offline Handwritten Text Recognition Based on CTC-Attention[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210007 Copy Citation Text

show less

Fig. 1. MTL framework based on CTC-Attention

Download full size

Fig. 2. CNN+LSTM structure

Download full size

Fig. 3. Example of original dataset

Download full size

Fig. 4. Word accuracy curves of different λ

Download full size

Model	CER(valid)	WER(valid)
CTC	9.7	27.6
Attention	7.1	19.1
MTL(λ=0.2)	6.6	18.2
MTL(λ=0.5)	7.4	19.8
MTL(λ=0.8)	10.4	29.2

Table 1. CER and WER of MTL on valid dataset unit: %

Method	Author	Pre-processing	Lexicon	Language model	Pre-train	CER	WER
RNN+CTC	Mor et al^[19]Krishnan et al^[20]Stunner et al^[21]Wiginton et al^[22]	√	2.4 million	√√	Synthetic	6.344.776.07	20.9016.1913.3019.07
Attention	Bluche et al^[6]Sueiras et al^[1]	√			CTC	12.60 8.80	23.80
CTC_Attention	Ours					6.60	18.20

Table 2. Comparison of the recognition rate of several popular methods on the IAM dataset

Yangyang Ma, Bing Xiao. Offline Handwritten Text Recognition Based on CTC-Attention[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210007

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information