基于文字结构的自切分手写汉字文本识别方法

打开文本图片集

中图分类号：TP311.5 文献标志码：A

Self-segmentation Recognition Method for Handwritten Chinese Character Text Based on Character Structure

GU Yi-man，ZHANG Xiao-lei （School of Computer Science and Technology，Qingdao University，Qingdao 266071，China）

Abstract： Handwritten Chinese text has complex structures， diverse writing styles，and unclear character boundaries， making traditional segmentation-free recognition methods prone to misalignment and recognition errors. A handwritten Chinese text recognition method based on a segmentation-based recognition framework was proposed，and a Swin attention mechanism module with self-attention and a sliding window approach was integrated，along with the embedding of the self-information of radicals. Experimental results show that the accuracy of the proposed method achieves 94.07% for handwritten Chinese text recognition， outperforming the current common recognition SVTR method by a margin of 0.57%

Keywords： handwritten Chinese text recognition; attention mechanism; self-information of radicals;convolutional neural network

随着深度学习和互联网技术的快速发展，各种终端设备获取了众多的图像和视频等媒体信息，这些信息中包含了大量的文本，丰富了信息世界，但也面临如何有效地从中提取并利用文本信息等问题。（剩余9788字）