rlightner / tesseractdotnet

Automatically exported from code.google.com/p/tesseractdotnet
0 stars 0 forks source link

OCRWord is empty #10

Open GoogleCodeExporter opened 8 years ago

GoogleCodeExporter commented 8 years ago
What steps will reproduce the problem?
1. processor.Recognize(bmp) or rocessor.AnalyseLayout(bmp)

Tesseract 2 had the possibility to retrieve the confidence and position of each 
word in the OCRWord class. I noticed when testing with the latest version that 
this class is empty and doesn't contains a word and the confidence is always 0 
even when i OCRed a image upsidedown.

How can i acces these values?

Original issue reported on code.google.com by tim.verv...@gmail.com on 12 Jul 2011 at 12:14

GoogleCodeExporter commented 8 years ago
What a coincidence!

We ran into the same issue while trying to convert our project from the 
previous Tesseract version to this new release. Unfortunately, we can't access 
the 'Confidence' variable on the Word class; it always returns 0.0. Previously, 
we used it to verify each word after doing text recognition with Tesseract.

Do we have to set 'Confidence' manually by using the method 
'UpdateConfidenceAndInsertTo()'? This one doesn't seem to impact the Confidence 
property, because the CharList variable has no items... So the problem seems to 
be related to the unsetted CharList.

So in short: why is the CharList property not set for a Word object? Because of 
this, the Confidence of a Word can't be used.

Original comment by arn...@gmail.com on 12 Jul 2011 at 1:07

GoogleCodeExporter commented 8 years ago
And the other variables (line, format, font, size, value) except for the 
co-ords are all 0. r552 works fine - although it didn't have the 
doc/block/para/line structure. Sample code:

        Dim doc As DocumentLayout = _ocrProcessor.AnalyseLayout(bmp)
        For Each blk As Block In doc.Blocks
            For Each para As Paragraph In blk.Paragraphs
                For Each ln As TextLine In para.Lines
                    For Each wrd As Word In ln.Words
                        For Each ch As Character In wrd.CharList
                            If ch.Value > 31 Then
                                sbText.Append(Chr(ch.Value))
                             End If
                        Next
                        sbText.Append(" ")
                    Next
                    sbText.Append(vbCrLf)
                Next
            Next
        Next

Original comment by hbeanl...@gmail.com on 13 Jul 2011 at 2:42

GoogleCodeExporter commented 8 years ago
AnalyseLayout detect blobs location only, I don't collect any information.
In my opinion, AnalyseLayout will be pre-processing step to identify
the location of blobs in which we can add more image processing
function to improve and pass them to tesseract parallel recognition
with other flag.

TesseractProcessor::Recognize enables to user to recognize text with
some parameters:
- PageSegmentMode: single block/para/textline/word.... for your purpose...
- OcrEngineMode: recognize with specific engine
- ROI, UseROI: recognize only inside a ROI

Goodluck!

Original comment by congnguy...@gmail.com on 13 Jul 2011 at 6:29

GoogleCodeExporter commented 8 years ago
Thanks you for your rapid response. I tried to use your settings instruction 
but I havent got any result with it. I am asking myself if it is possible to 
get the position of a char in combination with the char itself. If it is 
possible than maybe u can explain me how i can do this beside parsing the text 
i get from Recognize.

If this function is not implemented in the current version(590) then I am 
curious  if this function is begin implemented in the next version?

Thank you in advance.

Original comment by tim.verv...@gmail.com on 13 Jul 2011 at 8:18

GoogleCodeExporter commented 8 years ago
Try to search around in origin tesseract-ocr engine for "chopper"...
If you find any relative variables, you can use 
"TesseractProcessor::SetVariable(..)" function...

Sorry, I have no plan to next version now. I will announce about it ASAP!

Original comment by congnguy...@gmail.com on 13 Jul 2011 at 11:01

GoogleCodeExporter commented 8 years ago
I look forward to seeing this re-implemented, seeing as it was in r552, so it 
should not be hard to re-add.
This info is useful to reconstruct the document's layout (in rtf/html format). 
Tesseract gives us the co-ords, font class, font size which can be useful.

If a form is rubber stamped, i do not know exactly where the stamp is, so i 
should be able to look for some text of a certain format in a certain size and 
confidently identify it.

Original comment by hbeanl...@gmail.com on 26 Jul 2011 at 12:31

GoogleCodeExporter commented 8 years ago
Yeah please add this back in - desperately need confidence levels for license 
plate recognition

Original comment by ad...@developerinabox.com on 2 Sep 2011 at 5:11

GoogleCodeExporter commented 8 years ago
Using tesseractdotnetwrapper_r590.zip 
in every Sample*.cs and every input image I get only 1 block at
doc = processor.AnalyseLayout(bmp);
Is it tesseract.dll foult or missed configuration of processor?

VS2010 Debug x86
Win7 32-bit

Original comment by povver...@gmail.com on 14 Oct 2011 at 1:39