Optical Character Recognition (OCR) With Tesseract3

[jpbro]

I needed to perform OCR on values selected out of a larger image by users one-at-a-time, and that led me to Tesseract 3 (there are more recent versions, but I found V3 first and it worked well enough for me, so I just stuck with it).

I wrote a small class that makes it easy to pass an image of some text to Tesseract and have it spit back a VB6 String (shared it at the bottom of this post).

Some Notes:

• The class works with CDECL build of Tesseract 3 - I'm using the last 3.x series compiled and published by the University of Mannheim Library (UB Mannheim) that is available here: https://digi.bib.uni-mannheim.de/tesseract/ - Get the "tesseract-ocr-setup-3.05.02-20180621.exe" release.
  
  
• To make CDECL calls & Image handling easier, I'm using RC6 - feel free to swap it out for your favourite image handling & CDECL calling code if you aren't an RC6 user. If you aren't going to use RC6, please check the comments anyway as I ran into a few "gotchas" - reading them might save you a few hours of headaches.
  
  
• This class implements a very small subset of the Tesseract Base API. I only wrote what I needed to get the job done. Feel free to extend it as needed, and post your changes back here if you are so inclined.
  
  
• It's a bit slow - I only get around 15 image->text conversions per second against short input, but since my primary use-case is to convert $ values extracted from a user drawn region (one $ value at a time captured from a larger document with many values), it is more than fast enough.
  
  
• Tesseract3 seems to do best with high resolution images, especially for detecting the difference between small things like commas and periods - so I recommend working with 200-300DPI image sources if possible.

Public Methods of CTesseract3

Initialize - This must be called before you call any other methods - pass the path to the folder where libtesseract-3.dll is located.

AcceptableCharacters - Pass a string of characters that Tesseract3 will consider "legal" - this can improve accuracy if (for example) you want to detect only numbers, you can pass "0123456789" and it won't mis-detect things like "I", "O", etc... This is completely optional though.

GetStringFromImage - Pass an image file path or byte array (JPG/PNG/BMP) and it will spit back a String of the OCR'd image.

GetCurrencyFromImage - Pass an image file path or byte array (JPG/PNG/BMP) and it will spit back a Currency datatype of the OCR'd image. Use this method when you know you want to recognize Currency value images.

Basic Example Code:

Public Sub Test
   Dim tess As New CTesseract3

   tess.Initialize "<path to folder where libtesseract-3.dll is>"

   MsgBox tess.GetTextFromImage("<path to JPG/BMP/PNG image>") ' OR MsgBox tess.GetTextFromImage(<image byte array>)
End Sub

I've included a few small sample images in the project, and you can try out the OCR by typing "Test" in the Immediate window and press return - you should see the following if everything is working OK:

test
STATUS: PASS  GetStringFromImage aginst Text Image Result: This is a test of the emergency broadcast system.    Expecting: This is a test of the emergency broadcast system.
STATUS: FAIL  GetStringFromImage against Positive Currency Image Result: 7.086.74   Expecting: 7,086.74
STATUS: PASS  GetCurrencyFromImage against Positive Currency Image Result: 7086.74  Expecting: 7086.74
STATUS: PASS  GetCurrencyFromImage against Negative Currency Image Result: -7086.74 Expecting: -7086.74

(Note that the second item FAILing is the expected result - it's to illustrate the importance of using GetCurrencyFromImage for numeric value processing)

Feel free to ask questions/report any bugs, and I hope someone out there finds this useful.

Source Code:

Tesseract3a.zip