How to extend lines in binary image?

HI all,

I used an edge detector on a graysclae image and got a binary image that has many lines (lines are 1's and the background is 0's). Example is given bellow. I need to extend these lines, so that the intersections will devide the image to regions (marked in red).

I can use imdilate, but I dont want to make these lines thicker, I need to extend them. In some cases I need to extend these lines segnificantly, so imdilate cant do the job.

To conclude, I am looking for a way to extend binary lines (longer, not thicker) by a specified factor.

Thank you for your help.

Answer 1

[Image Analyst]

Try imclose(). It does imdilate() followed by imerode() to shring it back to the original size.

[Ilya Tcenov]

Thank you. It solved most of my problem. Although, there are cases that require line extension that cannot be done with imclose().

I am still looking for a way to extend lines (in some cases 1.5 times longer).

[Image Analyst]

If the lines you want to extend are all vertical, you can make a structuing element that is only vertical and very long, like a hundred pixels, and very thin horizontally, like 1 or 2 pixels.

[Ilya Tcenov]

Thank you for your suggestion.

Unfortunately, the lines can appear in all angles, positions and lengths.

[Dillon Hudson]

I'm also looking for a solution to this problem. I'm looking for a way to connect the boundary of a line to the next closest line (not necessarily another boundary). Perhaps based on the rough direction the line was pointed i.e. take a boundary point and add to that line along the path it would take if it was lengthed. My lines are only a pixel wide (from edge detection) and span a number of angles and lengths.

The picture attached below is after one iteration of imclose with a square structuring element. Should I be using a different structuring element?

Something like this would be acceptable where the pink lines represent what I'd like to be white.

[Image Analyst]

@Dillon Hudson please start your own discussion thread for this. It is possible though not trivial. And where the lines extend to depends on the order of the lines you connected before. It's not a unique solution.

Answer 2

[Víctor Llamas Martínez]

Hello.

In my case I ussually have to join points and lines that are closed together and in the most of the cases this code helps me out

se = strel('disk',2); 
bw = imclose(bw,se);
imshow(bw)

With this you create a disk around each point of your binnary image and you try to find close points that are not connected with it. There is an example:

In this figure the line that must be continius is in some points disconnected, but aplying those code lines everithing gets connected propertly:

Perhaps some artifacts can appear in your image but by tuning the disk radius i think you can archieve an acceptable result.