Variations of AlexNet and GoogLeNet to Improve Korean Character Recognition Performance
0
조회
0
다운로드
2
인용
0
참고문헌
키워드
초록
Deep learning using convolutional neural networks (CNNs) is being studied in various fields of imagerecognition and these studies show excellent performance. In this paper, we compare the performance ofCNN architectures, KCR-AlexNet and KCR-GoogLeNet. The experimental data used in this paper is obtainedfrom PHD08, a large-scale Korean character database. It has 2,187 samples of each Korean character with2,350 Korean character classes for a total of 5,139,450 data samples. In the training results, KCR-AlexNetshowed an accuracy of over 98% for the top-1 test and KCR-GoogLeNet showed an accuracy of over 99% forthe top-1 test after the final training iteration. We made an additional Korean character dataset with fonts thatwere not in PHD08 to compare the classification success rate with commercial optical character recognition(OCR) programs and ensure the objectivity of the experiment. While the commercial OCR programs showed66.95% to 83.16% classification success rates, KCR-AlexNet and KCR-GoogLeNet showed averageclassification success rates of 90.12% and 89.14%, respectively, which are higher than the commercial OCRprograms’ rates. Considering the time factor, KCR-AlexNet was faster than KCR-GoogLeNet when they weretrained using PHD08; otherwise, KCR-GoogLeNet had a faster classification speed.
Abstract
Deep learning using convolutional neural networks (CNNs) is being studied in various fields of imagerecognition and these studies show excellent performance. In this paper, we compare the performance ofCNN architectures, KCR-AlexNet and KCR-GoogLeNet. The experimental data used in this paper is obtainedfrom PHD08, a large-scale Korean character database. It has 2,187 samples of each Korean character with2,350 Korean character classes for a total of 5,139,450 data samples. In the training results, KCR-AlexNetshowed an accuracy of over 98% for the top-1 test and KCR-GoogLeNet showed an accuracy of over 99% forthe top-1 test after the final training iteration. We made an additional Korean character dataset with fonts thatwere not in PHD08 to compare the classification success rate with commercial optical character recognition(OCR) programs and ensure the objectivity of the experiment. While the commercial OCR programs showed66.95% to 83.16% classification success rates, KCR-AlexNet and KCR-GoogLeNet showed averageclassification success rates of 90.12% and 89.14%, respectively, which are higher than the commercial OCRprograms’ rates. Considering the time factor, KCR-AlexNet was faster than KCR-GoogLeNet when they weretrained using PHD08; otherwise, KCR-GoogLeNet had a faster classification speed.
참고문헌 (26)
- D. Cireşan, "Multi-column deep neural network for traffic sign classification", Neural Networks, 32, pp.333~338 (2012)[DOI]
- "A convolutional neural network for modelling sentences" (2014)[URL]
- P. L. Callet, "A convolutional neural network approach for objective video quality assessment", IEEE Transactions on Neural Networks, 17(5), pp.1316~1327 (2006)[DOI]
- D. C. Ciresan, "Convolutional neural network committees for handwritten character classification", Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), pp.1135~1139 (2011)
- T. Wang, "End-to-end text recognition with convolutional neural networks", Proceedings of the 21st International Conference on Pattern Recognition(ICPR 2012), pp.3304~3308 (2012)
- "Deep convolutional network for handwritten Chinese character recognition"[URL]
- Z. Zhong, "High performance offline handwritten Chinese character recognition using GoogLeNet and directional feature map", Proceedings of the 13th International Conference on Document Analysis and Recognition (ICDAR), pp.846~850 (2015)
- W. Yang, "Improved deep convolutional neural network for online handwritten Chinese character recognition using domain-specific knowledge", Proceedings of the 13th International Conference on Document Analysis and Recognition (ICDAR), pp.551~555 (2015)
- D. C. Hwang, "Hangul recognition using path following algorithm", IE Interfaces, 3(2), pp.53~62 (1990)
- B. K. Sin, "On-line handwritten character recognition with hidden Markov models", Proceedings of the 4th Annual Conference on Human and Cognitive Language Technology, pp.533~542 (1992)
- J. K. Chung, "A study on an on-line handwritten Hangul character recognition by identifying relative positions of strokes", Journal of Information Technology Applications and Management, 4(2), pp.65~778 (1997)
- J. Y. Ha, "Optimization of number of states in HMM for on-line Hangul recognition", Proceeding of the Korea Information Science Society, 25(2), pp.372~374 (1998)
- J. H. Lee, "Elastic curvature matching for online handwritten Hangul recognition", Proceeding of the Korea Information Science Society, 35(2), pp.238~239 (2008)
- H. S. Cho, "A new feature-extraction method using wavelet transformation and fuzzy data for handwritten Hangul recognition", Journal of Korean Institute of Information Technology, 3(4), pp.11~17 (2005)
- I. J. Kim, "Handwritten Hangul recognition using deep convolutional neural network", International Journal of Document Analysis and Recognition, 18(1), pp.1~13 (2011)
- 함대성, "한글 문자 데이터베이스 PHD08 구축", 한국콘텐츠학회 논문지, 8(11), pp.33~40 (2008), 한국콘텐츠학회
- Y. LeCun, "Gradient-based learning applied to document recognition", Proceeding of the IEEE, 86(11), pp.2278~2324 (1998)[DOI]
- A. Krizhevsky, "ImageNet classification with deep convolutional neural network", Proceedings of the 25th International Conference on Neural Information Processing Systems (NIPS'12), pp.1097~1105 (2012)
- "ImageNet Large Scale Visual Recognition Challenge"[URL]
- C. Szegedy, "Going deeper with convolutions", Proceeding of the IEEE Conference on Computer Vison and Patter Recognition, pp.1~9 (2015)
- "Provable bounds for learning some deep representations" (2013)[URL]
- "Linear interpolation"[URL]
- "Caffe (convolutional architecture for fast feature embedding)"[URL]
- "ABBYY FineReader 12"[URL]
- "ABC OCR scanner app"[URL]
- "Office Lens app"[URL]