Handwriting recognition in Tibetan based on active learning strategy

作者：Lili Wang, Sen Bao, Yanshan He, Hongwu Yang · 年份：2022 · DOI：10.1109/ialp57159.2022.9961323 · 被引用次数：5 · 研究领域：Handwritten Text Recognition Techniques、Image Retrieval and Classification Techniques、Hand Gesture Recognition Systems

Automatically recognizing Tibetan handwriting characters and digits has significant meaning for Tibetan information processing. This paper implemented a closed-loop feedback system of collection-training-recognition-retraining to recognize Tibetan handwriting by considering the lack of Tibetan handwritten sample data, inconsistent data formats, granularity, and other problems. A semi-automatic labeling system was proposed based on an active learning strategy to overcome the difficulty in data collection for small sample data and enhance the accuracy of the recognition model of Tibetan handwritten characters and digits. The recognition models, which utilized an improved LeNet-5 network structure, achieved 98.30% accuracy in handwritten digit recognition. Through this system, the model's accuracy will continue to improve after continuous iteration.