I am working on OCR system for Devanagari and trying to extract the basic building components. The basic components may include vertical lines, horizontal lines, curves, closed loops etc. As a segmentation result I want to extract these components. A sample of segmentation is shown in the figure attached here.
We call these kind of features (structural features), so there isn't a single way to compute them.
You may use the skeleton of that character in order to find the intersection points. Then you can analyze the segments between these points to see if it is a vertical/horizontal or curved line (see eccentricity).
To find loops, you may apply some processing in order to fill the holes and then substruct the filled image with the original, this will give your only the loops (see imfill matlab function)
Good luck
Dear sir
Just take the image of character and continue
Dear Mr. Nirajan Pant,
I can suggest you use the method of continuous skeletons. The result of applying this approach is shown in the picture. Polygonal border approximation is a green line, inscribed circles are a gray line, middle axes are a red line. You can find strokes and their geometric characteristics from the analysis of the skeleton graph.
Mestetskiy, L., Semenov, A. Binary image skeleton - continuous approach // VISAPP 2008 - 3rd International Conference on Computer Vision Theory and Applications, Proceedings 1, pp. 251-258
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing