Not able to extract table contents for Scanned pdf's and Normal Pdf's using Tesseract-ocr?

[Sayali begampure]

We are trying to extract text content from normal pdf and scanned pdf (image) using tesseract-ocr.

We have observed following issues for the pdf's with table as table Contents are not getting extracted properly.

1. Contents from few cells(rows/columns) are not visible.Sometimes heading of the table is missing.
2. If numbers are there inside table, all the numbers are not getting extracted.
3. Some letters are extracted wrongly . eg. i is misinterpreted as l.
4. Column sequence is getting interchanged as it is parsing horizontally.
5. Some extra characters are getting extracted along with normal one.

Tried image_to_string ,image_to_data ,opencv approach

Sample code used is:

from PIL import Image

import pytesseract from pytesseract import image_to_string from pytesseract import image_to_boxes

image=(pytesseract.image_to_string(Image.open('table_number.jpg'))) print(image)

It should extract rows and columns properly which it is not extracting as of now. Kindly suggest function or method to enhance the results for table content extraction using tesseract.

[Manasi sarode]

That's fair enough.

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/6ddfa19c-8025-40f8-8f17-a393e5b5b2cc%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Amulya Kali]

Did you try changing psm?

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAJ7g%3DmSUBC4aK0L%3De-9bbtBX5%3DCiFF%3DkLW8Wcmvr4YjQG13pmQ%40mail.gmail.com.

[Sayali begampure]

Used psm for 2 column documents. Its showing results perfectly.

Can you send link or pointers how to use it for table content extraction from scanned pdf?

Thanks

On Friday, 31 May 2019 16:00:36 UTC+5:30, Amulya Kali wrote:

Did you try changing psm?

On Fri, May 31, 2019, 15:57 Manasi sarode <manasi....@gmail.com> wrote:

That's fair enough.

On Fri, May 31, 2019, 3:55 PM Sayali begampure <sayalisb...@gmail.com> wrote:

We are trying to extract text content from normal pdf and scanned pdf (image) using tesseract-ocr.

We have observed following issues for the pdf's with table as table Contents are not getting extracted properly.

1. Contents from few cells(rows/columns) are not visible.Sometimes heading of the table is missing.
2. If numbers are there inside table, all the numbers are not getting extracted.
3. Some letters are extracted wrongly . eg. i is misinterpreted as l.
4. Column sequence is getting interchanged as it is parsing horizontally.
5. Some extra characters are getting extracted along with normal one.

Tried image_to_string ,image_to_data ,opencv approach

Sample code used is:

from PIL import Image

import pytesseract from pytesseract import image_to_string from pytesseract import image_to_boxes

image=(pytesseract.image_to_string(Image.open('table_number.jpg'))) print(image)

It should extract rows and columns properly which it is not extracting as of now. Kindly suggest function or method to enhance the results for table content extraction using tesseract.

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.

To unsubscribe from this group and stop receiving emails from it, send an email to tesser...@googlegroups.com.

To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/6ddfa19c-8025-40f8-8f17-a393e5b5b2cc%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.

To unsubscribe from this group and stop receiving emails from it, send an email to tesser...@googlegroups.com.

[Amulya Kali]

Hi,  I'm not sure about the psm mode you have used. You can try psm 6 for table. 

Something like this.. 

pytesseract.image_to_string(image, lang='eng', config='--psm 6')

To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.

To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/b9c108dc-141c-4eb1-8bea-654410e42e05%40googlegroups.com.

[Sayali begampure]

Thanks..I will try with this.

[Sayali begampure]

Hello ,I tried with both psm 6 and psm 3 , but still problem in detecting the table contents.Numbers are not visible and also sometimes only heading is visible.

Any other change I can do in tesseract or for image quality improvement?

TIA

[Krishna Prasad]

Hi Sayali,

     I'm dealing with a similar problem. Detecting table contents accurately has never been easy with tesseract. I would suggest building your own pipeline for Detecting tables and complex layouts. There are many public datasets available. I'm trying to use Deeplab V3+ by Google

Model : https://github.com/tensorflow/models/tree/master/research/deeplab

Dataset : https://www.primaresearch.org/datasets/Layout_Analysis

Deeplab is properly documented and really good at its job. If you are familiar with ML, this would be a piece of cake for you. 

Hope this helps. 😃

Regards,

Krishna Prasad A S

To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.

To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/b1816f57-a23f-4cad-b8a2-686fa40364bf%40googlegroups.com.

[Saylee Begampure]

Thanks a lot! Yes.. I am familiar with ML part, will implement it and try to get results.

[Akhil Dixit]

I am also facing same issue for Scan PDF specially with multiple columns and Text with numbers. Please share some inputs here if anyone tried using tesseract or some other APIs.