CODEWITHSUNDEEP
androidocrtesseract
HomeIsWhereThePcIs
HomeIsWhereThePcIs

Reputation: 1464

Parse only a specific part of image with Tesseract

I am trying to use Tesseract OCR on Android to read the state of a gas meter when you take a picture of it: picture

This is the output when I parse this image:

vb"
22% BK-G4T ||||||||I||||I|||ii\|||\
’ 64 2007
22?: 06.0"! 'm'lm Mm. 23212274 ,
v 2,0 dm’ 1
pmn 0_5 bar tm ~25°C v‘40"(1 I
1amp é 0_o1m’ sb15°cl :Sp 20°c l
'I ELSTEQ~I¢¢>>InstrogwnSs HB Z _ 18 _ 1013 . ‘
a, 069373593435- 3 I
i'23212214 Y _ w w V'
g

The idea is to extract the first 5 digits of the state of the gas meter ( 06937 on this image ).

My question is, is there a way to train Tesseract to only parse this part of the image? Absolute coordinates are not an option since every picture would be different. I am guessing the best logic would be something like: parse only white numbers on black background.

Upvotes: 1

Views: 4499

Answers (1)

thewaywewere
thewaywewere

Reputation: 8626

By changing the page segmentation mode (psm), tesseract 4.00.00 alpha is able to read the meter line characters correctly as 06937598-m3 apart from other characters.

The command used is:

tesseract meter.jpg output --psm 11 -l eng

--psm 11 means to recognize "Sparse text. Find as much text as possible in no particular order".

Here is the output file with showing all the ASCII control characters.

enter image description here

If --psm 11 works on other meter images, then you could just need to search -m3 at the end of the line to extract the who meter line characters. With that, you can get the first 5 digits right away.

Hope this help.

Upvotes: 2

Related Questions