serverside ocr

[cvlug]

Does anyone know how to use ocr (Optical Character Recognition)
with php?
I would like to translate text in an image into plain text with php.

If its not possible with php, is there an other way to do this serverside?

Thanks,

Charles.

[ednark]

well php wasn't built to have that in included...

however there are two options i see

1) find out how much fine grain control you can have over an image with the [man]image[/man] library that does come with php... if that gives you pixel by pixel control/reading of a file, then all you need to do is find the algorithm for character recognition and write it yourself... however it doesn't seem like you are up for that, nor would php be anyones language of choice to do that in..

2) php has command line access to [man]exec[/man] programs... by that i mean you can find some program that does do character regonition, have php ask it to run on the file you want, return the results to php and then use them however you want

this seems the easiest to do, if you have a program that runs on a filename and returns the basic text... if not, then whip open your book on shell scripting and write a wrapper than will take the parameters you want, and return the data you want...

[cvlug]

Thanks a lot for the very complete answer.
I found a sulotion as well.

It's possible to schedule scansoft omnipage to ocr the content of a directory, and write the text in an other directory.
With php it's possible to fill a directory with images to scan, and after omnipage has done its ocr, read the directory with textfiles.
However, ocr of .pdf files gives a poor result.

It must be possible to have a better result, because the html versions of pdf files that google providers in its searchresults are almost perfect. I am wondering how google reads pdf. Does anyone know?

Reading text from pdf files is possible with ghostscript. This is done from the commandline, and does not look very comfortable to use.

Does anybody knows an php script/class to read pdf?

Thanks,

Charles

[dcalloway]

PDF files can be produced by scanning a hardcopy version of a printed document, or by exporting from a word processor-type application. Files indexed by search engines come from the latter method as the ascii text is embedded in the PDF file. Text from scanned images does not get indexed.

[Turing_Machine]

I have found that running the PDF file through a Ghostscript FCGI process, then pushing the results through a print-to-tiff (mgetty, etc, treat it as a fax) will create a tiff file that can then be OCR'd the normal way. This way, you get all the text in the images, etc. Ghostscript does break sometimes, but for standard PDF files, it works great.