I would look at the first derivative of the color in the image. That should give you every point in the image with drastic color change (i.e. a border). Then it's a question of connect-the dots. If you have a cartoonish image you might get the problem that black outlines of figures will show up twice. To solve that, you could use an algorithm that merges similar lines?

Is this supposed to be something that works on-the-fly in real-time video or just for static images?

Use one of the edge detectors and run them over the image. This should help you produce a binary image, which essentially is a 2D array.

OpenCV has pretty powerful built-in functions for contour finding and management. A simple example on the usage of these is given here:

http://docs.opencv.org/doc/tutorials/imgproc/shapedescriptors/find_contours/find_contours.html

I encourage you to look through the documentation of the functions.

Morphological operators can be an interesting direction to pursue as well. You can find a lot of material on morphological operators online, for example you can find a very brief introduction (mostly images) here:

http://www.vlsi.uwindsor.ca/presentations/2007/13-Neil.pdf

One of the main advantages of morphological operators is that they are very simple to implement.

@Dominik Steenken

Thanks for your interest in this. I would have line-only drawing as pixels (eg. png file), black-and-white. It does not need to do it in real-time (as if processing a video stream).

@Anuj Sehgal

Which edge detectors do you refer to? Any C++ library that does edge detection?

@Marek Kraft

Thanks for suggestion, I've used OpenCV before as base for facial tracking, however it was (to my recall) quite robust solution and tons of code that I don't need for this purpose. Of course if I don't find any other way, will have to dig into OpenCV source to extract what I need. The app I am to develop must be lightweight and will be running on a server side, thus must be very low on resource usage.

I was thinking if there are known algorithms of tracing edges/outlines from pixel array, and then create implementation of such algorithm in C++ to test it out.

@Marco Trincavelli

Thanks for the tip, I am reading it through now at it looks quite promising! Thanks!

@Marco Trincavelli

I've read through pdf and it shows techniques of how to create a derivate image that will show edges of the source image. Cool theory however I would rather need different output, in form of vector data, for example:

http://www.visionmax.com.hk/tmp/201212071038.png

@Subrata Ashe

Thanks for your links, unfortunately these are all purchase-only articles and from screen shots they offer it looks to me it's still having image as the end result, and not numerical data, do I understand it correctly?

@Subrata Ashe

Thanks, I will try to get these papers through some friends first, but many thanks on your kind offer to work on details! Let me first try to see if I can crunch this on my own, if not would you mind if I contact you through email?

If you are looking at doing something simple, then you don't even need to do edge detection, especially if you are sure you will have lines on white paper. Simply put, what you can do is use thresholds to convert the image into a binary image.

As far as edge detectors go, for your purpose, I don't have a specific suggestion. Use any that you can find in C++ libraries, or use OpenCV since that might help you do this faster. If none of this works, let me know and I can give you a matrix that works as a simple edge detector if you convolute an image with it.