I have a PDF with a big (multi page) table and I'm trying to extract all text present in that table, using the QPDF library for Delphi 2007 and ExtractFilePageText function (with Options "8" as the best algorithm choice for that type of table).

The problem is: not all table cells are filled with text , some are empty, so I'm going crazy searching a way to let the library put in the output text file some sort of placeholder to inform me that some cells are empty. Now I have an output text file with a new line of text for every table cell with text into, but NO lines at all when a cell contains no text (so no way to know that in a specific line and column there is an empty cell, a type of information that I need to recostruct the whole table content).

Marco

you should use option 4.

Then you'll get returning strings with position data.

Having the position data page by page you can check where there's something missing.

My idea is getting data for example for row 1 col 1, row 1 col 3, row 1 col 4, row 2 col 1, row 2 col 2, ... let you know that row 1 col 2 is an empty field ;-)

Cheers and welcome here,

Ingo

 

I've take a look a bit deeper to the option 4.

First of all I think I've found a little problem in the documentation: the coordinates of the points returned in the option 4 are in the Y,X order, not X,Y (moving from a cell of the table to the next one in horizontal order - same row - it seems that are the Ys that are changing instead of, of course, the Xs.

That said, yes, did you tell me to look for the coordinates returned cell by cell and from those values, with a simple algorithm, find if there is some cell that is missing from the extraction process? thank you, I'll try this path now!

Marco

 

regarding the coordinates this could have to do with the use of rotate-functionality. Rotation will rotate the position data, too ;-)

If you want to calculate and using some algos you should Keep in mind that pdf-documents can look similar but can be different inside. For example two documents with format DIN A4 but one was initially landscape and then rotated to DIN A4. So check the rotation first (there's an explained function in the reference) and use SetOrigin.

 

Cheers, Ingo

You should call QP.NormalizePage(0); before calling the text extraction functions.

A PDF file has no concept of cells, sentences or paragraphs.  PDF text is just placed on the page like a jigsaw puzzle.  Some pieces can contain 1 character, others multiple characters and there are many functions that make this more complex such as character spacing, word spacing etc..

Debenu Quick PDF Library just like any other PDF library has to collect all the characters and their positions on the page and group characters into words and then into lines.  It is not an exact science.

Your best option is to try to use GetPageText(7) and the SetTextExtractionScaling function if you are using tightly spaced proportional fonts.

If you provide a link to the PDF I could then suggest the best method for extraction.  I have a few years of experience in complex table extraction from PDF files.

Andrew

This is the link to that PDF file: 

http://www.inbet.it/pdf/calcioq.pdf

As you see, some cells are empty (without odds), so my problems.

I've followed the path suggested by Ingo, with some improvement (but the algorithm is not trivial ;) ).

Marco

There is going to be no easy way to read this with Debenu Quick PDF Library without some complex strategies.

How many pages do you need to process and how often ?

Andrew.

my files are not a lot of pages long, but I need to process them ...20 times a day, every day of the year.

So, according to Ingo suggestions I've now developed some Delphi code with the option 4, using x and y coordinates to find if a cell is empty (so, skipped by the library) and the results are not so bad. Anyway, I'll continue to search for a better solution, thank you all.

Try this

  QP.SetTextExtractionScaling(0, 2, 8);

  QP.SetTextExtractionWordGap(0.2);

And then call GetPageText(7);  You can then split the text into columns and trim leading and trailing spaces.

Andrew.