Legacy ImageMagick Discussions Archive

I'm looking for a way to automatically remove text from an image.

I've seen a javascript project that can do it ( http://projectnaptha.com/ )

It uses OCR and then an inpaint method.

I'm curious is there is another way, specifically, server side.

inpainting has been requested as a future enhancement to IM. User snibgo has developed a method for doing it with IM Magick Filter. See his hole filling section at his web site at http://im.snibgo.com/ and also viewtopic.php?f=1&t=28640#p127233.

Also see the use of morphology for small text at viewtopic.php?p=41498#p41498

Yes, once text is deleted (by which I mean "the pixels are made transparent"), it can be inpainted. I show a number of techniques.

But first, the pixels forming the text must be identified. The Naptha page says, "The primary feature of Project Naptha is actually the text detection, rather than optical character recognition", and cites a Microsoft paper. When the object is to remove text, this is the important step. It has occurred to me that implementing the described detection method might be easy to implement in IM.