Best Practices and Performance Optimization Solutions
According to the project documents, to achieve peak recognition rate of 95% need to follow three principles: input quality recommended 300DPI or more scanned documents, fuzzy images will reduce the detection rate of chart elements; parameter configuration should be enabled -dpi parameter to match the resolution of the source document, complex documents are recommended to add -verbose Log analysis of the source of the error; API selection of mathematical formulas preferred to use MathPix, multi-language tables recommended to use Google Vision. a typical case is the processing of mathematical papers containing fractional matrices, the use of the combination of -mode math + -dpi 600 can be the formula to identify correct rates from 82% to 93%. The project also provides -compress parameter to optimize the output of large files, 10,000 pages of PDF compressed JSON volume can be reduced by 65%.
This answer comes from the articleVOP: OCR Tool for Extracting Complex Diagrams and Math FormulasThe
English 
简体中文 
日本語 
Deutsch 
Português do Brasil