计算机科学 ›› 2020, Vol. 47 ›› Issue (8): 221-226.doi: 10.11896/jsjkx.190500017

• 计算机图形学&多媒体 • 上一篇    下一篇

基于残差连接的场景文本识别端到端网络结构优化

黄金星, 潘翔, 郑河荣   

  1. 浙江工业大学计算机科学与技术学院 杭州 310023
  • 出版日期:2020-08-15 发布日期:2020-08-10
  • 通讯作者: 郑河荣(hailiang@zjut.edu.cn)
  • 作者简介:981667807@qq.com
  • 基金资助:
    国家自然科学基金(61871350)

End-to-end Network Structure Optimization of Scene Text Recognition Based on Residual Connection

HUANG Jin-xing, PAN Xiang, ZHENG He-rong   

  1. College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou 310023, China
  • Online:2020-08-15 Published:2020-08-10
  • About author:ZHENG He-rong, born in 1971, Ph.D, professor, supervisor.His main research interests include pattern recognition and image processing.
  • Supported by:
    This work was supported by the National Natural Science Foundation of China (61871350).

摘要: 针对已有文本识别网络由于深度不够而识别准确率较低的问题, 文中提出一种改进的端到端文本识别网络结构。首先, 将文本作为序列, 采用残差模块将文本按列切分成特征向量输入循环层。这种残差结构增加了卷积网络的深度, 使网络保持对文本图像的最佳表征能力, 实现对文本信息的捕捉。另一方面, 残差模块采用堆叠层来学习残差映射, 在层数加深的情况下提高了网络的收敛性。然后, 采用循环层对这些文本特征序列进行上下文建模, 并把建模结果输入Softmax层以获得序列对应标签的预测, 实现了对任意长度文本的识别。循环层使用长短时记忆网络学习文本之间的依赖关系, 解决长序列训练过程中的“梯度消失”问题。最后, 通过最优路径方法进行文本标签转录。该方法找到一条路径使其概率最大, 并输出这条路径对应的序列为最优序列。改进的文本识别网络结构增加了深度, 提高了文本图像的特征描述能力和在噪声下的稳定性。在多个测试数据集(ICDAR2003, ICDAR2013, SVT和IIIT5K)上将所提算法与已有典型算法进行实验对比分析, 结果表明该网络结构能够得到更高的场景文本识别准确率, 验证了其有效性。

关键词: 残差连接, 场景文本识别, 堆叠层, 网络深度, 最优路径

Abstract: The existing text recognition methods will cause decreased recognition accuracy due to not enough network depth.The paper addresses this issue and proposes an improved end-to-end text recognition network structure.Firstly, the algorithm takes the text as a sequence, and uses the residual module to divide the text into columns for the recurrent layer.This residual structureincreases network depth, to maintain the network’s best representation of the text image.It can capture the best feature representation of text images.Meanwhile, the residual module uses the stacked layer to learn the residual mapping to improve the convergence of the network though the number of layers is obviously increased.Secondly, we use the recurrent layer to model the context of these text features, and the modeling results will be taken into the softmax layer to predict corresponding labels, which achieve the recognition of arbitrary length of texts.The recurrent layer uses the Long Short-Term Memory to learn the dependencies between texts and solve the gradient vanishing problem in long sequence training.Finally, text label transcription and decoding are performed by the optimal path method.The method finds a path to maximize its probability, and outputs the sequence corresponding to the path as the optimal sequence.The improved text recognition network structure increases network depth, improves the feature description of text images and the stability under noises.In the experimental part, this paper compares with existing typical algorithms over the multiple test datasets (ICDAR2003, ICDAR2013, SVT and IIIT5K).The experiments show that the network structure can obtain better text recognition accuracy and verify the effectiveness of the proposed network structure.

Key words: Network depth, Optimal path, Residual connection, Scene text recognition, Stacked layer

中图分类号: 

  • TP311
[1]SHI B G, YANG M K, WANG X G, et al.ASTER:An Attentional Scene Text Recognizer with Flexible Rectification[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(9):2035-2048.
[2]SHI B G, BAI X, YAO C.An end-to-end trainable neural net-work for image-based sequence recognition and its application to scene text recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(11):2298-2304.
[3]CHENG Z Z, BAI F, XU Y L, et al.Focusing attention:Towards accurate text recognition in natural images[C]∥Proceedings of the IEEE International Conference on Computer Vision.2017:5086-5094.
[4]LEE C Y, OSINDERO S.Recursive recurrent nets with attention modeling for ocr in the wild[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:2231-2239.
[5]ALSHARIF O, PINEAU J.End-to-end text recognition withhybrid HMM maxout models[J].arXiv:1310.1811, 2013.
[6]SIMONYAN K, ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[J].arXiv:1409.1556, 2014.
[7]JADERBERG M, SIMONYAN K, VEDALDI A, et al.Synthetic data and artificial neural networks for natural scene text recognition[J].arXiv:1406.2227, 2014
[8]HE K M, ZHANG X Y, REN S Q, et al.Deep residual learning for image recognition[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:770-778.
[9]BAI X, YANG M K, SHI B G, et al.Scene text detection and recognition based on deep learning [J].Scientia Sinica Informationis, 2018, 48(5):531-544.
[10]WANG K, BABENKO B, BELONGIE S.End-to-end scene text recognition[C]∥2011 International Conference on Computer Vision.IEEE, 2011:1457-1464.
[11]DALAL N, TRIGGS B.Histograms of oriented gradients forhuman detection[C]∥International Conference on Computer Vision & Pattern Recognition (CVPR’05).IEEE Computer Society, 2005:886-893.
[12]YAO C, BAI X, SHI B, et al.Strokelets:A learned multi-scale representation for scene text recognition[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2014:4042-4049.
[13]BISSACCO A, CUMMINS M, NETZER Y, et al.Photoocr:Reading text in uncontrolled conditions[C]∥Proceedings of the IEEE International Conference on Computer Vision.2013:785-792.
[14]GOODFELLOW I J, WARDE-FARLEY D, MIRZA M, et al.Maxout networks[J].arXiv:1302.4389, 2013.
[15]GOODFELLOW I J, BULATOV Y, IBARZ J, et al.Multi-digit number recognition from street view imagery using deep convolutional neural networks[J].arXiv:1312.6082, 2013.
[16]JADERBERG M, SIMONYAN K, VEDALDI A, et al.Deepstructured output learning for unconstrained text recognition[J].arXiv:1412.5903, 2014.
[17]LECUN Y, BOTTOU L, BENGIO Y, et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE, 1998, 86(11):2278-2324.
[18]SU B, LU S.Accurate scene text recognition based on recurrent neural network[C]∥Asian Conference on Computer Vision.Cham:Springer, 2014:35-48.
[19]GRAVES A, FERNNDEZ S, GOMEZ F, et al.Connectionist temporal classification:labelling unsegmented sequence data with recurrent neural networks[C]∥Proceedings of the 23rd International Conference on Machine Learning.ACM, 2006:369-376.
[20]BAEK J, KIM G, LEE J, et al.What is wrong with scene text recognition model comparisons? dataset and model analysis[J].arXiv:1904.01906, 2019.
[21]BORISYUK F, GORDO A, SIVAKUMAR V.Rosetta:Largescale system for text detection and recognition in images[C]∥Proceedings of the 24th ACM SIGKDD International Confe-rence on Knowledge Discovery & Data Mining.ACM, 2018:71-79.
[22]LUO C, JIN L, SUN Z.Moran:A multi-object rectified attention network for scene text recognition[J].Pattern Recognition, 2019, 90:109-118.
[23]IOFFE S, SZEGEDY C.Batch normalization:Accelerating deep network training by reducing internal covariate shift[J].arXiv:1502.03167, 2015.
[24]GUPTA A, VEDALDI A, ZISSERMAN A.Synthetic data for text localisation in natural images[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:2315-2324.
[25]LUCAS S M, PANARETOS A, SOSA L, et al.ICDAR 2003 robust reading competitions[C]∥Seventh International Confe-rence on Document Analysis and Recognition.IEEE, 2003:682-687.
[26]KARATZAS D, SHAFAIT F, UCHIDA S, et al.ICDAR 2013robust reading competition[C]∥2013 12th International Conference on Document Analysis and Recognition.IEEE, 2013:1484-1493.
[27]MISHRA A, ALAHARI K, JAWAHAR C.Scene text recognition using higher order language priors[C]∥Proceedings of British Machine Vision Conference (BMVC).2012:1-11.
[28]KINGMA D P, BA J.Adam:A method for stochastic optimization[J].arXiv:1412.6980, 2014.
[29]WANG J, HU X.Gated recurrent convolution neural network for ocr[C]∥Advances in Neural Information Processing Systems.2017:335-344.
[30]CHENG Z Z, XU Y L, BAI F, et al.Aon:Towards arbitrarily-oriented text recognition[C]∥Proceedings of the IEEE Confe-rence on Computer Vision and Pattern Recognition.2018:5571-5579.
HUANG Jin-xing, born in 1995, postgraduate.Her main research interests include pattern recognition and image processing.
[1] 单美静, 秦龙飞, 张会兵.
L-YOLO:适用于车载边缘计算的实时交通标识检测模型
L-YOLO:Real Time Traffic Sign Detection Model for Vehicle Edge Computing
计算机科学, 2021, 48(1): 89-95. https://doi.org/10.11896/jsjkx.200800034
[2] 刘俊, 徐平平, 武贵路, 彭杰.
室内环境下基于最优路径规划的PSO-ACO融合算法
PSO-ACO Fusion Algorithm Based on Optimal Path Planning in Indoor Environment
计算机科学, 2018, 45(11A): 97-100.
[3] 张燕.
基于混沌优化的最优运输路径问题研究
Research on Optimal Transportation Route Based on Chaos Optimization
计算机科学, 2017, 44(Z6): 133-135. https://doi.org/10.11896/j.issn.1002-137X.2017.6A.030
[4] 方贤文,陶小燕,刘祥伟.
基于模态Petri网的行为有效区间寻找最优路径的方法
Method of Optimal Path Selection Based on Modal Petri Net Branching Effective Range
计算机科学, 2014, 41(7): 91-96. https://doi.org/10.11896/j.issn.1002-137X.2014.07.018
[5] 张倩,王振哗,董亚非.
分子算法在公交网络问题中的应用
Application of the Molecular Algorithm in Public Transport Network Problem
计算机科学, 2012, 39(2): 262-267.
[6] .
PGA:一种基于最优路径的Ad Hoc网络地理路由算法

计算机科学, 2008, 35(2): 19-22.
[7] .
层次化移动管理中的最优路径选择问题

计算机科学, 2006, 33(8): 42-45.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!