A handwritten image dataset comprising over 1 000 writers, covering 10 000 Mongolian characters, and totaling 300 000 samples, was constructed. Based on this dataset, this paper proposed a lightweight recognition model named MLCNet. The model used MobileNetV3 as the backbone network and introduced an efficient channel attention module in the high-level feature extraction stage to enhance the representation capability of key features. It employed a bidirectional long short-term memory network for temporal modeling and realized sequence-to-text mapping via connectionist temporal classification. Validation results on 60,000 test samples show that MLCNet achieves a recognition accuracy of 91.57%, only 2.47% lower than the baseline model using ResNet as the feature extraction network, while the parameter count is approximately 1/44 of that of the baseline model, significantly improving deployment feasibility on resource-constrained devices. The constructed dataset and lightweight model provide a feasible reference solution for the task of handwritten Mongolian character recognition.
HEPZIJ L, MUTHUMANII, SELVABHARATHIS. English cursive hand written character recognition[J]. Advances in Natural and Applied Sciences, 2017, 11(7): 57-64.
[2]
GUPTAJ D, SAMANTAS, CHANDAB. Ensemble classifier‐based off‐line handwritten word recognition system in holistic approach[J]. IET Image Processing, 2018, 12(8): 1467-1474.
[3]
BHAGYASREEP V, JAMESA, SARAVANANC. A proposed framework for recognition of handwritten cursive English characters using DAG-CNN[C]//Proceedings of the 1st International Conference on Innovations in Information and Communication Technology (ICIICT). Chennai: IEEE, 2019: 1-4.
[4]
WIGINGTONC, STEWARTS, DAVISB, et al. Data augmentation for recognition of handwritten words and lines using a CNN-LSTM network[C]//14th IAPR International Conference on Document Analysis and Recognition (ICDAR). Kyoto: IEEE, 2017: 639-645.
[5]
JADERBERGM, SIMONYANK, ZISSERMANA. Spatial transformer networks[C]//Advances in Neural Information Processing Systems 28 (NeurIPS 2015). Montréal: Curran Associates, Inc., 2015: 2017-2025.
[6]
DUTTAK, KRISHNANP, MATHEWM, et al. Improving CNN-RNN hybrid networks for handwriting recognition[C]//16th International Conference on Frontiers in Handwriting Recognition (ICFHR). Niagara Falls: IEEE, 2018: 80-85.
[7]
ALMAZÁNJ, GORDOA, FORNÉSA, et al. Word spotting and recognition with embedded attributes[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(12): 2552-2566.
[8]
KRISHNANP, JAWAHARC V. Generating synthetic data for text recognition[J]. arXiv Preprint arXiv, 2016:1608.04224.
[9]
VINCIARELLIA, LUETTINJ. A new normalization technique for cursive handwritten words[J]. Pattern Recognition Letters, 2001, 22(9): 1043-1050.
[10]
YANGH L, JINL W, HUANGW G, et al. Dense and tight detection of Chinese characters in historical documents: Datasets and a recognition guided detector[J]. IEEE Access, 2018, 6: 30174-30183.
[11]
YANGW X, JINL W, LIUM F. DeepWriterID: An end-to-end online text-independent writer identification system[J]. IEEE Intelligent Systems, 2016, 31(2): 45-53.
[12]
ZHANGX Y, BENGIOY, LIUC L. Online and offline handwritten Chinese character recognition: A comprehensive study and new benchmark[J]. Pattern Recognition, 2017, 61: 348-360.
[13]
WANGW C, ZHANGJ S, DUJ, et al. DenseRAN for offline handwritten Chinese character recognition [C]//16th International Conference on Frontiers in Handwriting Recognition (ICFHR). Niagara Falls: IEEE, 2018: 104-109.
[14]
MELNYKP, YOUZ Q, LIK Q. A high-performance CNN method for offline handwritten Chinese character recognition and visualization[J]. Soft Computing, 2020, 24(11): 7977-7987.
[15]
KRICHENO, CORBILLÉS, ANQUETILÉ, et al. Combination of explicit segmentation with Seq2Seq recognition for fine analysis of children handwriting[J]. International Journal on Document Analysis and Recognition (IJDAR), 2022, 25(4): 339-350.
[16]
ELLEUCHM, JRABAS, KHERALLAHM. The effectiveness of transfer learning for Arabic handwriting recognition using deep CNN[J]. Journal of Information Assurance & Security, 2021, 16(2):85-93.
[17]
ELLEUCHM, KHERALLAHM. Convolutional deep learning network for handwritten Arabic script recognition[M]//Hybrid Intelligent Systems. Cham: Springer, 2020: 103-112.
[18]
AMROUCHM, RABIM, ES-SAADYY. Convolutional feature learning and CNN based HMM for Arabic handwriting recognition[M]//Image and Signal Processing. Cham: Springer, 2018: 265-274.
[19]
RAWLSS, CAOH G, KUMARS, et al. Combining convolutional neural networks and LSTMs for segmentation-free OCR[C]//14th IAPR International Conference on Document Analysis and Recognition (ICDAR). Kyoto: IEEE, 2017: 155-160.
[20]
KUNDUS, PAULS, KUMAR BERAS, et al. Text-line extraction from handwritten document images using GAN[J]. Expert Systems with Applications, 2020, 140: 112916.
[21]
CHAKRABORTYA, DER, MALAKARS, et al. Handwritten digit string recognition using deep autoencoder based segmentation and ResNet based recognition approach[C]//25th International Conference on Pattern Recognition (ICPR). Milan: IEEE, 2020: 7737-7742.
[22]
NOUBIGHZ, MEZGHANIA, KHERALLAHM. Densely connected layer to improve VGGnet-based CRNN for Arabic handwriting text line recognition[J]. International Journal of Hybrid Intelligent Systems, 2022, 17(3-4): 113-127.
[23]
ZHANGH, WEIH X, BAOF L, et al. Segmentation-free printed traditional Mongolian OCR using sequence to sequence with attention model[C]//14th IAPR International Conference on Document Analysis and Recognition (ICDAR). Kyoto: IEEE, 2017: 585-590.
[24]
CUIS D, SUY L, QINGDAOERJIR, et al. An end-to-end network for irregular printed Mongolian recognition[J]. International Journal on Document Analysis and Recognition (IJDAR), 2022, 25(1): 41-50.
[25]
MAL L, JIL, WUJ. A new database for online handwritten Mongolian word recognition[C]//23rd International Conference on Pattern Recognition (ICPR). Cancun: IEEE, 2016: 1131-1136.
LIUJ, MAL L, WUJ. Online handwritten Mongolian word recognition using a novel sliding window method with recurrent neural networks[C]//14th IAPR International Conference on Document Analysis and Recognition (ICDAR). Kyoto: IEEE, 2017: 189-194.
[28]
WEIH X, LIUC, ZHANGH, et al. End-to-end model for offline handwritten Mongolian word recognition[M]//Natural Language Processing and Chinese Computing. Cham: Springer, 2019: 220-230.
[29]
滕达. 基于融合模型的蒙古文联机手写识别研究[D]. 呼和浩特: 内蒙古大学, 2023.
[30]
PANY C, FAND, WUH J, et al. A new dataset for Mongolian online handwritten recognition[J]. Scientific Reports, 2023, 13(1): 26.
[31]
李敏. 基于深度学习的联机蒙古文手写识别系统研究[D]. 呼和浩特: 内蒙古大学, 2019.
[32]
潘月彩. 蒙古文联机手写生成和识别技术研究[D]. 呼和浩特: 内蒙古大学, 2023.
[33]
FAND, SUNY X, WANGZ X, et al. Online Mongolian handwriting recognition based on encoder⁃decoder structure with language model[J]. Electronics, 2023, 12(20): 4194.
[34]
BAEKY, NAM D, PARKS, et al. CLEval: Character-level evaluation for text detection and recognition tasks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Seattle: IEEE, 2020: 564-565.
[35]
SHAFIEEM J, LIF, CHWYLB, et al. Squishednets: Squishing SqueezeNet further for edge device scenarios via deep evolutionary synthesis[J]. arXiv Preprint arXiv,2017:1711.07459.
[36]
RENQ D E J, WANGL L, MAZ R, et al. Offline Mongolian handwriting recognition based on data augmentation and improved ECA-net[J]. Electronics, 2024, 13(5): 835.