Online Handwritten Mongolian Character Recognition using CMA-MOHR and Coordinate Processing

摘要

With the objective of improving the performance of online handwritten traditional Mongolian recognition, this paper proposes a segmentation-free, lightweight online handwritten Mongolian character recognition model network structure CMAMOHR and coordinate processing methods applied to online handwritten Mongolian recognition jobs. A new method for resampling coordinate sequences was used in the preprocessing. We also built a large dataset of Mongolian online handwriting. Compared to the baseline model, our model had a 28.94% and 15.65% improvement in whole word recognition topl and toplO, respectively. Our method enables higher recognition rates and faster convergence.