The project code is mainly used to generate OCR related text training data. The project contains a large number of font files, corpus files, near-shaped characters, etc., of which there are more than 20 fonts, corpus data up to 2G, including company corpus Address corpus, novel corpus to meet various needs.
The data generated by the project is used for the training of multiple identification models such as ID cards, business cards, and bills. Comprehensive Synthetic Chinese String Dataset extraction code: fh6h, training recognition The accuracy of the model can reach more than 99%. The text generated by the project is very similar to the 360w data, so it can supplement the insufficient data in the 360w data. The project contains functions such as image rotation, perspective transformation, enhancement, etc. It is powerful, flexible and diverse! Because the corpus files and background images are too large, corpus data network disk address extraction code: awfn
The selected background picture is stored in the back_ground folder in the project. The background picture processing can use the text.py file in the back_ground to enhance the background to meet the needs of recognizing the scene.
The font_file /font_all/ directory in the project stores font files. New fonts need to run the check_font.py file to extract font fonts. If the corresponding file in font_in_all is blank, the font may have errors in use. Results in a blank.
The text_file folder in the project stores corpus data, and the newly added corpus text can be placed in this directory.
Imagaug_image.py in the project can be added in various forms. For details, please refer to the imagaug library
Configure the path in the main.py file, run the main.py file to generate it, and modify the word.py to increase the text sequence of the required scene.