A method and device for automatically detecting and recognizing mathematical formulas in a complex scene
By combining the YOLOv1 object detection model and the LaTeX-OCR recognition model, the problems of difficulty in locating mathematical formula regions and insufficient recognition accuracy in complex scenarios are solved, and stable and editable LaTeX expression generation is achieved, which is suitable for online education and intelligent grading systems.
Patent Information
- Application Number
- CN202610287987.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-10
- Publication Date
- 2026-07-17
AI Technical Summary
Existing mathematical formula recognition systems struggle to accurately locate formula regions in complex scenarios, and their recognition accuracy is insufficient, failing to maintain structural hierarchical information and resulting in chaotic recognition results.
The YOLOv1 object detection model is used to accurately extract the formula region, and the LaTeX-OCR recognition model is combined for structured recognition. The original image data is processed by the pre-trained YOLOv1 object detection model to obtain the predicted bounding box of the mathematical formula and crop it. Then, the LaTeX-OCR recognition model is used for feature encoding and sequence decoding to generate the LaTeX expression.
It achieves stable, editable and highly accurate mathematical formula recognition in complex scenarios, improves recognition accuracy and interoperability between systems, and is applicable to various types of image scenarios, especially in the presence of interference factors, it can still accurately locate the formula area.
Smart Images

Figure CN122416474A_ABST