improve tesseract OCR number recognition

1
votes

I've been playing around with tesseract and the different psm options, im trying to work with this pattern:

enter image description here

When I process this I get 52658, it switches up the 5's and the 8's and losses the decimal place. Is there anyway I could get a more precise reading on this:

The Image was initially transparent i added the white background with PHP to try to give it better recognition, no result.

1
phpocrtesseract
You could try to preprocess the image and erode it. – rr-

1 Answers

4
votes

The image is simply too small.

I resized it using ImageMagick and it started to OCR correctly, with Tesseract both 3.02 and 3.03:

$ tesseract 8UAYy.png ooo
Tesseract Open Source OCR Engine v3.03 with Leptonica
$ cat ooo.txt 
B2 655

$ convert 8UAYy.png -resize 300% ooo.png
$ tesseract ooo.png ooo
Tesseract Open Source OCR Engine v3.03 with Leptonica
$ cat ooo.txt 
82.685

$ tesseract302 ooo.png ooo
Tesseract Open Source OCR Engine v3.02.02 with Leptonica
$ cat ooo.txt 
82.685