Language-sensitive multilingual text recognition method based on attention mechanism
By constructing a language-sensitive multilingual text recognition model, the problem of poor recognition results caused by large differences in the length of multilingual text images is solved, and high-precision recognition of texts in different languages is achieved, especially with a significant improvement in the recognition results of short and long texts.
CN116343225BActive Publication Date: 2025-12-23HEFEI INSTITUTE OF PHYSICAL SCIENCE CHINESE ACADEMY OF SCIENCES
-1 Cites -1 Cited by
Patent Information
- Application Number
- CN202211737696.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-30
- Publication Date
- 2025-12-23
- Estimated Expiration
- 2042-12-30
Smart Images
Figure CN116343225B_ABST
Abstract
The application discloses a kind of language-sensitive multilingual text recognition methods based on attention mechanism, its steps include: first, collect text picture with annotation, establish text recognition dataset;Then construct language-sensitive multilingual text recognition model based on attention mechanism;Then run language-sensitive multilingual text recognition model based on attention mechanism;Finally, the loss function corresponding to each text picture is calculated, and the language-sensitive multilingual text recognition network based on attention mechanism is trained;Finally, the trained network is used to identify the picture, and the recognition result is obtained.The application does not need to train the respective recognition network for different languages, and can be targeted to different languages for recognition, so as to realize balanced high-precision recognition on text images in each language, to improve the recognition rate.
Need to check novelty before this filing date? Find Prior Art