Programmatically recognize text from scans in a PDF File


I have a PDF file, which contains data that we need to import into a database. The files seem to be pdf scans of printed alphanumeric text. Looks like 10 pt. Times New Roman.

Are there any tools or components that can will allow me to recognize and parse this text?

Best Solution

I've used pdftohtml to successfully strip tables out of PDF into CSV. It's based on Xpdf, which is a more general purpose tool, that includes pdftotext. I just wrap it as a Process.Start call from C#.

If you're looking for something a little more DIY, there's the iTextSharp library - a port of Java's iText - and PDFBox (yes, it says Java - but they have a .NET version by way of IKVM.NET). Here's some CodeProject articles on using iTextSharp and PDFBox from C#.

And, if you're really a masochist, you could call into Adobe's PDF IFilter with COM interop. The IFilter specs is pretty simple, but I would guess that the interop overhead would be significant.

Edit: After re-reading the question and subsequent answers, it's become clear that the OP is dealing with images in his PDF. In that case, you'll need to extract the images (the PDF libraries above are able to do that fairly easily) and run it through an OCR engine.

I've used MODI interactively before, with decent results. It's COM, so calling it from C# via interop is also doable and pretty simple:

' lifted from
Dim inputFile As String = "C:\test\multipage.tif"
Dim strRecText As String = ""
Dim Doc1 As MODI.Document

Doc1 = New MODI.Document
Doc1.OCR()  ' this will ocr all pages of a multi-page tiff file
Doc1.Save() ' this will save the deskewed reoriented images, and the OCR text, back to the inputFile

For imageCounter As Integer = 0 To (Doc1.Images.Count - 1) ' work your way through each page of results
   strRecText &= Doc1.Images(imageCounter).Layout.Text    ' this puts the ocr results into a string

File.AppendAllText("C:\test\testmodi.txt", strRecText)     ' write the OCR file out to disk

Doc1.Close() ' clean up
Doc1 = Nothing

Others like Tesseract, but I have direct experience with it. I've heard both good and bad things about it, so I imagine it greatly depends on your source quality.

Related Question