Result of OCRTesseractDoOCR dont match with GetCharCount

[Slava]

Hi,

I suspect a bug in somewhere in the OCR Tesseract plugin introduced since GdPicture version 5.12.0/5.12.1 (?)

OCRTesseractGetCharCount returns for some documents a larger number than length of a string (without spaces/new lines) returned from OCRTesseractDoOCR function. If I do the following loop:

Code: Select all

for i:=1 to oGdPicture.OCRTesseractGetCharCount do
  teststring := teststring + Char(oGdPicture.OCRTesseractGetCharCode(i));

Then I compare this teststring to the Result from OCRTesseractDoOCR and there are 2 more chars placed in the middle of text! At position 46 and 51 in my current test document these appear: #25 and ¬
They do not exist in the string returned from OCRTesseractDoOCR function.

Please respond as soon as possible.

Kind regards,
Slava

[Loïc]

Hi,

Please send us code snippet + original document to reproduce this error.

Kind regards,

Loïc

[Slava]

Hi Loic,

I've created a code snippet for you:

Code: Select all

var
  i, j: Integer;
  OCRResultNoSpaces, OCRResult, teststring: String;
begin
  oGdPicture.SetNativeImage(viewer.GetNativeImage);
  OCRResult := oGdPicture.OCRTesseractDoOCR(TesseractDictionaryDutch, ocrDictionaryPATH, '');
  OCRResultNoSpaces := StringReplace(OCRResult, ' ', '', [rfReplaceAll]);
  OCRResultNoSpaces := StringReplace(OCRResultNoSpaces, #13, '', [rfReplaceAll]);
  OCRResultNoSpaces := StringReplace(OCRResultNoSpaces, #10, '', [rfReplaceAll]);

  j := 1;
  for i:=1 to oGdPicture.OCRTesseractGetCharCount do
    begin
      teststring := teststring + Char(oGdPicture.OCRTesseractGetCharCode(i));
      if Char(oGdPicture.OCRTesseractGetCharCode(i)) <> OCRResultNoSpaces[j] then
        ShowMessage('!Foreign Char @ pos '+ IntToStr(i) +': '+ Char(oGdPicture.OCRTesseractGetCharCode(i)))
      else
        Inc(j);
    end;
  //breakpoint here and evaluate OCRResultNoSpaces vs teststring
  Showmessage('Same results: '+ BoolToStr(teststring = OCRResultNoSpaces, True));
end;

About 50% of all invoices produce this behavior. I will send a few as soon as I can to esupport address.

Kind regards,
Slava

[Loïc]

Hi Slava,

Problem solved (was an unicode issue).

I will publish a fixed edition on Monday.

Kind regards,

Loîc

[Slava]

Hi Loic,

When do you plan to release the new version? It's tuesday now and there's still no fix released. At least not for the ActiveX version.

Kind regards,
Slava

[Loïc]

Hi,

I will be able to publish the new edition at the end of the week due to H5N1 (I must stay at home)...

Thank you for your comprehension,

Loïc

[Slava]

Loic,

I apologize for rushing things up, I didn't know that. It is a dangerous type, are you going to be ok?

Good luck with it.

Slava

[Loïc]

Thank you Slava !

It is not dangerous at all. I am just a bit tired. All will be ok within 1 day or two with new release for .NET & ActiveX editions !

Cheers,

Loïc

[JRaboin]

Good day,

It seems that this problem exists in version 10.2.0.30 of GdPicture.

From what I can see <spaces> are not included in the OCRTesseractGetCharCount call. Along with that \r and/or \n are sometimes included. The recognition seems random but it is difficult to get an accurate count/confidence per character when reading thru the string returned form the OCRTesseractDoOCR call.

int[] coords = StringToInts(fldData[2]);
m_GdPictureImaging.SetROI(coords[0] + offset[1], coords[1] + offset[2], coords[2], coords[3]);
string tmpRetval = m_GdPictureImaging.OCRTesseractDoOCR(imageID, _dictionary, _dictionaryPath, _charWhiteList);

// Get characters read and each characters confidence
//
_OCRcharCount = m_GdPictureImaging.OCRTesseractGetCharCount();
//

The length of tmpRetval when recognizing a column of information is always larger then the value of _OCRcharCount.

Please advise.

[Loïc]

Hello,

We are already handling your issue in the helpdesk system. So please do not cross post here.

We will give you a feedback on the helpdesk as soon we will have something.

With best regards,

Loïc

[Loïc]

btw the answer was:

Spaces are not considered as characters by the OCR engine.
If you want to get the number of spaces between 2 recognized glyphs just use the OCRTesseractGetCharSpaces() method.

Please let me know if you need further information.

[JRaboin]

Loïc,

Please accept my apology. I do not recall logging an issue on this particular problem. It was just a quick google search brought this topic up and I thought a quick post against it would help not only me but others if they experience the same issue.

Best regards,
James