Controllable Text Layout Generation For Synthesizing Scene Text Image

In this paper, we propose an algorithm for generating a controllable synthetic dataset of Chinese text layouts. This algorithm allows users to specify the arrangement direction, segmentation method, and curvature of the text, enabling the generation of more complex text layouts. Our algorithm provides flexible parameter control, allowing users to generate Chinese text datasets with diverse layouts. Additionally, we introduce the ControlNet model for style transfer, enriching the generated data with detailed information and backgrounds. Through comparative experiments and ablation experiments, our algorithm has been validated, demonstrating its effectiveness and superiority in generating controllable layouts for Chinese text datasets. The incorporation of the ControlNet model for style transfer enhances the data generation process by enriching the details and backgrounds. Consequently, the synthesized dataset generated by our algorithm provides a valuable resource for training and evaluating Chinese text analysis algorithms, further advancing research and development in the field. Our algorithm lays the groundwork for exploring new technologies and promoting innovation in the realm of Chinese text analysis.