Publications by authors named "Zecheng Xie"
Learning Spatial-Semantic Context with Fully Convolutional Recurrent Network for Online Handwritten Chinese Text Recognition.
IEEE Trans Pattern Anal Mach Intell
August 2018
Article Synopsis
- Online handwritten Chinese text recognition (OHCTR) poses challenges due to its large character set, ambiguous segmentation, and variable input lengths, necessitating advanced techniques.
- The paper introduces a method that uses path signature to create signature feature maps from pen-tip trajectories, enabling effective capture of pen stroke characteristics.
- The proposed multi-spatial-context fully convolutional recurrent network (MC-FCRN) improves predictions by leveraging various spatial context information while an implicit language model incorporates semantic understanding, leading to impressive accuracy rates of 97.50% and 96.58% on standard datasets.