P2H-GAN: An Effective Method For Generating Handwritten Expressions

Handwritten text generation, particularly for complex two-dimensional structured texts like mathematical expressions, remains a largely unexplored area in digital documentation. To fill this research gap, we introduce P2H-GAN, a novel GAN-based model specifically designed for generating both one-dimensional and intricate two-dimensional structured handwritten texts. This model innovatively integrates printed image information into the content encoder and utilizes distinct style and content encoders to capture calligraphic styles and multimodal content. To ensure the generation of realistic handwritten texts, we establish three complementary learning objectives. Comprehensive experiments have demonstrate that P2H-GAN not only excels in creating high-quality images of handwritten texts but also enhances the datasets for Handwritten Mathematical Expression Recognition (HMER), leading to improved recognition performance.