tessdata_ssd

Traineddata for Tesseract 4 for recognizing Seven Segment Display

This is a proof of concept traineddata in response to these posts in tesseract-ocr google group, 1 and 2.

The training text and scripts used are provided for reference.

Feel free to clone the repo and rerun training with your own custom training_text and fonts. You will also need files from tesseract_ocr/langdata and tesseract_ocr/tessdata_best.

Finetune Plus Training from tessdata_best/eng.traineddata

Finetune Plus training was done using a training text of 9700+ lines with only numbers and limited punctuation (. : /) etc.

Three Seven Segment Display fonts were used for making synthetic training data.

• "DB LCD Temp Heavy" \
• "SF Digital Readout Heavy" \
• "Seven Segment" \

14000 iterations - char train=0.186%, word train=0.4%

• Download best version - 10.7 MB. use with -l ssd.

• Download fast/integer version - 1.37 MB. use with -l ssd_int.

• Download older best version - 10.9 MB. use with -l 7seg.

Evaluation

Evaluation was done using synthetic data generated using an evaluation text of over 440 lines with only numbers and limited punctuation (. : /) using same 3 fonts.

Test of a small sample of real life images gives better results with the older 7seg.traineddata. See https://github.com/Shreeshrii/tessdata_ssd/issues/1 for details

Other SSD Projects (I have not tried them)

• https://www.unix-ag.uni-kl.de/~auerswal/ssocr/

• https://github.com/auerswal/ssocr

• https://github.com/arturaugusto/display_ocr