Layout Analysis using OCRopus

Here we provide an on-line demo of our layout analysis algorithm used in the OCRopus open source OCR system. For detailed information about the algorithm, please refer to:

T. M. Breuel: High Performance Document Layout Analysis, Symposium on Document Image Understanding Technology, Greenbelt, Maryland, 2003.

Notes:

This is a simple visualization of a debugging interface for the layout analysis. It is not a standalone layout analysis system. In particular, it makes a number of assumptions:
- The input is deskewed.
- The input has been denoised.
- Images in the input have been replaced with black boxes.
- The image is in a "normal" resolution range (around 200-400 dpi)
- The input is in a writing system like Latin, Japanese, Russian, etc. (the system can handle Urdu, Devanagari, etc. as well, but that requires some parameter changes to work well)
The method will give reasonable results on some inputs that violate some of its assumtions, but YMMV.
We're planning on updating the demo in the future to contain the necessary preprocessing, cleanup, and script detection; check in again in a couple of months.

You can either submit an image through the form interface, or you can submit it programmatically through HTTP. You can also submit a PDF document, in which case the first page will be rendered at 200dpi and then used.

Form Interface

If you do not have an image at hand or want to try some of our images, try one of these (note that results are cached, so this is faster than using a new image):

Programmatic Interface

To submit your image programmatically, you can simply POST to this URL; the image should be a parameter named "imagefile".

From the command line, you can do this using:

curl -D header.out -F 'imagefile=@input.jpg;type=image/jpeg' http://demo-madm.dfki.uni-kl.de/layout/ > output.png

You can also do this easily using the HTTP implementation in your favorite programming language (C#, Python, Java, Perl, etc.).