Each OCR engine had its own minimum text height, and none had a maximum up to 60 px
When a screenshot is too small for OCR, upscaling it first is the tempting move. After three years on video codecs that bothers me: the detail was spent, or not, at capture time. So here is a test of how much text height each recognizer needs, and what upscaling recovers below it. The samples are deliberately clean: a self-written Chinese passage of 275 characters and an English one of 452, black…
Each optical character recognition (OCR) engine had its own minimum text height requirement, with no upper limit beyond 60 pixels. Various tests were conducted to determine the minimum height needed for each recognizer to accurately interpret text. The experiments utilized a self-written Chinese passage of 275 characters and an English passage of 452 characters, rendered at nine different heights ranging from 6 to 60 pixels using two Chinese fonts (Songti SC and Heiti SC) and displayed on various platforms.
The accuracy of the OCR engines was measured by calculating 1 minus the character edit distance over the length, where one wrong Chinese character equated to 0.36 points. The results showed that the minimum height for accurate interpretation varied among the engines and fonts, with some engines performing better at certain heights than others.
For instance, macOS built-in text recognition in accurate mode was error-free on Chinese text at 12 pixels for Heiti and 10 pixels for Songti, while Tesseract English required 16 pixels for clean interpretation. The findings suggest that upsampling images to a larger text height can improve OCR accuracy, but the effectiveness depends on the specific engine and font being used.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
