Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced Datasets

In this paper, we propose a novel deep model for unbalanced distribution Character Recognition by employing focal loss based connectionist temporal classification (CTC) function. Previous works utilize Traditional CTC to compute prediction losses. However, some datasets may consist of extremely unba...

Full description

Saved in:
Bibliographic Details
Main Authors: Xinjie Feng, Hongxun Yao, Shengping Zhang
Format: Article
Language:English
Published: Wiley 2019-01-01
Series:Complexity
Online Access:http://dx.doi.org/10.1155/2019/9345861
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849404419749183488
author Xinjie Feng
Hongxun Yao
Shengping Zhang
author_facet Xinjie Feng
Hongxun Yao
Shengping Zhang
author_sort Xinjie Feng
collection DOAJ
description In this paper, we propose a novel deep model for unbalanced distribution Character Recognition by employing focal loss based connectionist temporal classification (CTC) function. Previous works utilize Traditional CTC to compute prediction losses. However, some datasets may consist of extremely unbalanced samples, such as Chinese. In other words, both training and testing sets contain large amounts of low-frequent samples. The low-frequent samples have very limited influence on the model during training. To solve this issue, we modify the traditional CTC by fusing focal loss with it and thus make the model attend to the low-frequent samples at training stage. In order to demonstrate the advantage of the proposed method, we conduct experiments on two types of datasets: synthetic and real image sequence datasets. The results on both datasets demonstrate that the proposed focal CTC loss function achieves desired performance on unbalanced datasets. Specifically, our method outperforms traditional CTC by 3 to 9 percentages in accuracy on average.
format Article
id doaj-art-8fc834f8bc5c4bb5ab5c93873dee3364
institution Kabale University
issn 1076-2787
1099-0526
language English
publishDate 2019-01-01
publisher Wiley
record_format Article
series Complexity
spelling doaj-art-8fc834f8bc5c4bb5ab5c93873dee33642025-08-20T03:36:59ZengWileyComplexity1076-27871099-05262019-01-01201910.1155/2019/93458619345861Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced DatasetsXinjie Feng0Hongxun Yao1Shengping Zhang2Harbin Institute of Technology, ChinaHarbin Institute of Technology, ChinaHarbin Institute of Technology, ChinaIn this paper, we propose a novel deep model for unbalanced distribution Character Recognition by employing focal loss based connectionist temporal classification (CTC) function. Previous works utilize Traditional CTC to compute prediction losses. However, some datasets may consist of extremely unbalanced samples, such as Chinese. In other words, both training and testing sets contain large amounts of low-frequent samples. The low-frequent samples have very limited influence on the model during training. To solve this issue, we modify the traditional CTC by fusing focal loss with it and thus make the model attend to the low-frequent samples at training stage. In order to demonstrate the advantage of the proposed method, we conduct experiments on two types of datasets: synthetic and real image sequence datasets. The results on both datasets demonstrate that the proposed focal CTC loss function achieves desired performance on unbalanced datasets. Specifically, our method outperforms traditional CTC by 3 to 9 percentages in accuracy on average.http://dx.doi.org/10.1155/2019/9345861
spellingShingle Xinjie Feng
Hongxun Yao
Shengping Zhang
Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced Datasets
Complexity
title Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced Datasets
title_full Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced Datasets
title_fullStr Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced Datasets
title_full_unstemmed Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced Datasets
title_short Focal CTC Loss for Chinese Optical Character Recognition on Unbalanced Datasets
title_sort focal ctc loss for chinese optical character recognition on unbalanced datasets
url http://dx.doi.org/10.1155/2019/9345861
work_keys_str_mv AT xinjiefeng focalctclossforchineseopticalcharacterrecognitiononunbalanceddatasets
AT hongxunyao focalctclossforchineseopticalcharacterrecognitiononunbalanceddatasets
AT shengpingzhang focalctclossforchineseopticalcharacterrecognitiononunbalanceddatasets