WordCanvas
This project is a text image rendering tool based on Pillow, designed specifically for random image generation.
By adding a variety of parameter settings, users can flexibly adjust input text, font styles, colors, and other attributes to achieve large-scale random generation of text images. Whether addressing issues such as data scarcity, class imbalance, or enhancing image diversity, WordCanvas provides a simple and efficient solution, offering a solid data foundation for deep learning training.
📄️ Introduction
In the current research on Optical Character Recognition (OCR), the accuracy of models depends on the quality and diversity of the dataset.
📄️ Installation
We offer installation via PyPI or by cloning this project from GitHub.
📄️ Quick Start
Starting is always the hardest part, so we need a simple beginning.
📄️ Advanced
In addition to basic usage, we also provide advanced settings to give you more flexibility in controlling the output text images. Here, we introduce randomization features, mainly used for training models.
📄️ Augmentation
We did not implement the image augmentation feature within WordCanvas because we consider it a highly "customized" requirement. Different use cases may require different augmentation methods. However, we provide some simple examples to demonstrate how the image augmentation process can be implemented.
📄️ MRZGenerator
After completing the development of WordCanvas, we can use this tool for other tasks as well.
📄️ BarcodeGenerator
This functionality is a small feature we often implement in practice. Creating a separate project for it would be excessive, so we included it here.
📄️ Resources
Text synthesis tools are primarily used for automatically generating image datasets, especially in scenarios where large amounts of annotated data are required to train deep learning models. These tools enhance model adaptability to different environments, fonts, colors, and backgrounds by embedding synthetic text in images to simulate real-world occurrences of text.