how to compute edge roughness for a binary region

I am new to the field of imaging processing. This may seem to be very trival to human eye and maybe very trival to people in this field as well. I am trying to classify two types of objects. They have very similar shape,but some of them have smooth edges and some of them have rather rought edge. I wonder how I can quantify the edge roughness, so that the computer will see this difference,and classify these two objects into two different groups? For example, to seperate a smooth 2D shape like a watermelon, and a similar 2D shape with some sparks on the edge.

Thanks for any help!

[Image Analyst]

Classify them HOW? Do you want to detect if they are the same shape, regardless if the edges/boundary may be smooth or rough (like a sharp triangle and a ragged one)? Or do you want to somehow characterize the roughness of the edge itself for a given shape? Those are two different things. Which do you want?

Answer 1

[Sean de Wolski]

Yes. You just have to define "smooth" and "rough" edges in terms that a computer can understand. I would recommend starting by defining them in English (or native language) and then trying to teach the computer what they are.

A smooth edge for my research (edges of glass beads) is probably much different than a smooth edge in your research. Post a few images and explain what you're looking for in order to receive less generalized assistance.

[cloudia]

Thanks for the reply!

I guess that is my question: how I can quantify edge roughness. For example, to seperate a smooth 2D shape like a watermelon, and a similar 2D shape with some sparks on the edge.

BTW, how do I upload images? I am sorry I am new here.

[Sean de Wolski]

http://www.mathworks.com/matlabcentral/answers/7924-where-can-i-upload-images-and-files-for-use-on-matlab-answers

First calculate the edge. The one way that comes to mind is to traverse it a few pixels at a time calculating the orientation of those pixels. Then calculate the angle between every few pixels' orientations. Angles greater than some threshold would constitute a rough transition.

Weekend is beginning here, but I can probably help you with coding early next week.

Good Luck!

[Sean de Wolski]

And Welcome to MATLAB Answers!

[cloudia]

That would be great! I appreciate it.

[Ashish Uthama]

cloudia, you can (re)edit the question to add new/clarifying information, helps people know what you are looking for without having to go through the comments :).

[cloudia]

Thanks for the suggestion! Just edit it,hopefully it will be less confusing.

Answer 2

[Image Analyst]

OK, your clarification says that you want to characterize edge roughness, or basically how wiggly and tortuous the perimeter is. Several ways to do this and no single right answer. A "right" answer is one that works for you with your particular images. You can use fractal dimension, circularity (perimeter^2 / (4*pi*Area)), Fourier Descriptors, and maybe some other things. Perhaps you can do some research here first: http://iris.usc.edu/Vision-Notes/bibliography/contents.html

Also, you haven't posted a link to your image(s) yet. You can do that if you want extra help.