ChineseOCR is an open-source optical character recognition project for detecting and reading Chinese text in images. It pairs a YOLOv3-based text detector with an OCR recognition stage, targeting real-world material such as documents, ID cards and train tickets. The project has about 6,100 stars on GitHub, is MIT licensed, and its codebase was last updated in 2022.
Reading Chinese text from photos is a two-part problem: first find where the text lines are, then recognize the characters. ChineseOCR tackles detection with a Darknet-style YOLOv3 model (with OpenCV DNN in the pipeline) and follows it with a recognition network, producing text strings from natural-scene and document images. The repository bundles the models and code needed to run the full pipeline yourself, which made it a popular self-hosted starting point for Chinese OCR before larger vision-language models arrived.
The repository is self-contained: model definitions, trained weights and inference code are all included, and the pipeline can be driven from Python for batch document processing. Its design reflects its era โ detection and recognition are separate specialized networks rather than one end-to-end vision-language model โ which keeps it fast and understandable on modest hardware, at the cost of accuracy on handwriting, heavily rotated text or low-quality scans compared with today’s much larger OCR systems. For clean printed Chinese documents, though, it remains a serviceable free option.
Free. ChineseOCR is open source under the MIT license and runs entirely on your own machines.
Developers who need self-hosted Chinese OCR for document-processing pipelines and want a small, understandable codebase rather than a cloud OCR API.