Legacy ImageMagick Discussions Archive

Here is an interesting paper I came acrosss:

http://www.bmva.org/bmvc/2012/BMVC/pape ... per063.pdf

It discusses the use of SWT (stroke width transform) with a novel "bandlet-based edge detector" in order to localize text occurances within images.

One of the frequently-asked questions on these forums is, "How can I clean up images like this one so an OCR can read it?"

The answer is usually very dependant on the nature of the image. We have seen scans and photographs of books and newspapers, but also camera-phone photos of till receipts. The methods in that paper may provide a generic solution.