Which method can we use to connect a skeleton road ?

I'm trying to determine all the possible paths to matches extremities of those curves (Road following)and choosing paths which verify a configuration like X, Y, T, _, and L in order to connect them(red curve) in a road skeleton not connected (white curve) like the example below :

I detected feature points like the picture below

but i'm stuck i don't know what approach use to do it because it requires a dynamique way .

i have to use a dynamic research window to determine the curve shape(configuration like X, Y, T, _, and L) that will complete the road network.

Any ideas please! Help is much appreciated

Answer 1

[Image Analyst]

I'm not sure what you're asking or how to answer. With the bwmorph() function you can find endpoints and branchpoints (the red * in your images). Any branchpoint can be checked out to see what it is. There are 256 possible combinations of the 8 surrounding pixels and there may be certain patterns that define each. Then you can make a look up table, like if it's

[0, 1, 0; 
 1, 1, 1;
 0, 1, 1]

it's an "X".

[mika]

Can you explain please your ideas with an example

the problem is on how to connect curves in a dynamic window.

[Image Analyst]

Well, what shape would you call the pattern example I gave? If the central pixel is set, you can have up to 4 separate "arms" branching off of it. If it has one arm, it's an endpoint or an elbow. If it has two arms, it's a line/curve. It if has 3 arms, it's a Y. If it has 4 arms it's an X. If you want to distinguish between Y and T you can do that. So just go through all 256 possibilities and make up a look up table.

[mika]

the problem is that i have to connect the white curves in which i complete the skeleton by the red curves that verify patterns(X, Y, T, _, and L) because initially they don't exist and i have to find a method that can make this . is it possible ?

any ideas please! thank you

[Image Analyst]

What do you mean by "connect" and what are "the red curves"? I don't see any red curves, just a set of isolated (x,y) coordinates where you placed a *. And like I said, you can construct a lookup table based on the pattern of the surrounding 8 pixels to determine if it's an X, a Y, etc.

[mika]

I'm talking about red curves mentioned in the image below that it's expected to be the result i'm looking for taking for example the L pattern mentioned by the blue circle

 can you explain a little bit how i can use a look up table because no idea about it and i don't know if it can be extended in a way to recover the more important feature points to give the right pattern.