Extract columns of text from a pdf file using iText

I need to extract text from pdf files using iText.

The problem is: some pdf files contain 2 columns and when I extract text I get a text file where columns are merged as the result (i.e. text from both columns in the same line)

this is the code:

public class pdf
{
    private static String INPUTFILE = "http://www.revuemedecinetropicale.com/TAP_519-522_-_AO_07151GT_Rasoamananjara__ao.pdf" ;
    private static String OUTPUTFILE = "c:/new3.pdf";

    public static void main(String[] args) throws DocumentException, IOException {
        Document document = new Document();
        PdfWriter writer = PdfWriter.getInstance(document, new FileOutputStream(OUTPUTFILE));
        document.open();

        PdfReader reader = new PdfReader(INPUTFILE);
        int n = reader.getNumberOfPages();

        PdfImportedPage page;

        // Go through all pages
        for (int i = 1; i <= n; i++) {
            page = writer.getImportedPage(reader, i);
            Image instance = Image.getInstance(page);
            document.add(instance);
        }

        document.close();

        PdfReader readerN = new PdfReader(OUTPUTFILE);
        for (int i = 1; i <= n; i++) {
            String myLine = PdfTextExtractor.getTextFromPage(readerN,i);
            System.out.println(myLine);

            try {             
                FileWriter fw = new FileWriter("c:/yo.txt",true);
                fw.write(myLine);
                fw.close();
            }catch (IOException ioe) {ioe.printStackTrace(); }
    }
}

Could you please help me with the task?

Solution

I am the author of the iText text extraction sub-system. What you need to do is develop your own text extraction strategy (if you look at how PdfTextExtractor.getTextFromPage is implemented, you will see that you can provide a pluggable strategy).

How you are going to determine where columns start and stop is entirely up to you - this is a difficult problem - PDF doesn't have any concept of columns (heck, it doesn't even have a concept of words - just putting together the text extraction that the default strategy provides is quite tricky). If you know in advanced where the columns are, then you can use a region filter on the text render listener callback (there is code in the iText library for doing this, and the latest version of the iText In Action book gives a detailed example).

If you need to obtain columns from arbitrary data, you've got some algorithm work ahead of you (if you get something working, I'd love to take a look). Some ideas on how to approach this:

Use an algorithm similar to that used in the default text extraction strategy (LocationAware...) to obtain a list of words and X/Y locations (be sure to account for rotation angle as well)
For each word, draw an imaginary line running the full height of the page. Scan for all other words that start at the same X position.
While scanning, also look for words that intersect the X position (but do not start on the X position). This will give you potential location for column start/stop Y positions on the page.
Once you have column X and Y, you can resort to a region filtered approach

Another approach that may be equally feasible would be to analyze draw operations and look for long horizontal and vertical lines (assuming the columns are demarcated in a table-like format). Right now, the iText content parser doesn't have callbacks for these operations, but it would be possible to add them without major difficulty.