Skip to content

Latest commit

 

History

History
31 lines (19 loc) · 801 Bytes

README.md

File metadata and controls

31 lines (19 loc) · 801 Bytes

Text Position Detector

Detects rectangular regions containing multilingual text in an image.

This project is based on Michael R. Lyu, Jiqiang Song and Min Cai's "A comprehensive method for multilingual video text detection, localization, and extraction". The extraction part is left out.

The default configuration works best for horizontally-aligned, well contrasted sinograms.

Usage

Run TextPositionDetector.apply to get a list of rectangles where text was detected.

Examples

Different parameters give different results depending on the characteristics of the input image. Here on inputs taken from video games:

  • Good result:

Good result output

  • Minor mishaps:

Minor mishaps output

  • Bad result:

Bad result output