Contrastive representation enhancement and learning for handwritten mathematical expression recognition

Zihao Lin, Jinrong Li, Gang Dai, Tianshui Chen, Shuangping Huang, Jianmin Lin

Abstract: Highlights•Contrastive learning based method learn robust symbol representation.•Printed expression symbols serve as semantic template of handwritten symbols.•Semantic-NCE loss applies on printed and handwritten expression by separating symbol semantic from writing appearance.•The proposed method achieves SOTA performance on benchmark datasets.