Finish lines in the image

Dear all,

I have an image like the one attached. I have it in binary format.

How do I complete the white lines so the first lines crosses x=0 point, and the remaining lines are finished until they cross each other?

This script should be flexible enough so the number of lines may change (more gaps) and they don't necessarily are lines. They may also be curves.

Here is another example where there are more gaps in data.

Thank you for your help!

[KALYAN ACHARJYA]

Are you want to link those white lines?

[Mateusz Matysiak]

Yes, I want to connect these lines

[Mateusz Matysiak]

I want that there is continuous white line from x=0 point till the last x value = over the entire graph

[Matt J]

Mateusz Matysiak's comment moved here:

The lines should look like below. Initially I was looking into creating multiple linear fit lines. Then, I create the complete lines based on the fits and cross points between the fits.

The picture below shows how it should look like. I finished the lines by hand. The script should do the same.

[Matt J]

This script should be flexible enough so the number of lines may change (more gaps) and they don't necessarily are lines. They may also be curves.

You mentioned here that the pieces might be curved (not lines) and so naive line-fitting wouldn't work. Please show an example of this more complicated case and how you want the curves extended.

Answer 1

[Matt J]

Here is another method based on interpolation, and faster than dilating with large strels.

[m,n]=size(A);

[I,J]=find(bwskel(A));

IJ=sortrows([I,J],2);

[I,J]=deal(IJ(:,1),IJ(:,2));

Jw=(J(1):J(end)).';
Iw=round(interp1(J,I,Jw));

B=imdilate(  accumarray([Iw,Jw],1,[m,n])   , strel('sphere',1));

imshow(B)

[Stelios Fanourakis]

Matt

Using your last suggestion I get the error

Error using griddedInterpolant
The grid vectors must contain unique points.

Error in interp1 (line 151)
        F = griddedInterpolant(X,V,method);

[Matt J]

Try this,

[m,n]=size(A);
[I,J]=find(bwskel(A));

[J,idx]=unique(J);
I=I(idx);

Jw=(J(1):J(end)).';
Iw=round(interp1(J,I,Jw));
B=imdilate(  accumarray([Iw,Jw],1,[m,n])   , strel('sphere',1));
imshow(B)

[Stelios Fanourakis]

I think I prefer your first suggestion even slower. This one I tried, and although it is faster indeed but I loose substantial information of my image.

[Matt J]

Here's a 3rd iteration of the interpolation based method (Method 2).

[m,n]=size(A);
[I,J]=find(A);
[G,J]=findgroups(J);
I=splitapply(@median, I,G);

Jw=(J(1):J(end)).';
Iw=round(interp1(J,I,Jw));
B=imdilate(  accumarray([Iw,Jw],1,[m,n])   , strel('sphere',1));

When I test this on lines2.mat, I get a much better completion than my original slower method.

Method 1:

Method 2:

[Stelios Fanourakis]

Still I struggle to get the correct results. Though I will stick to your first suggestion that works very good for me. Thanks Matt.

[Matt J]

You are welcome, but please Accept-click an answer if you have found one that works for you.

Answer 2

[Matt J]

Perhaps as follows,

 gapsize=36;
 
 A=imcrop(rgb2gray(imread('image.png')));
 
 B=(imdilate(A,strel('sphere',ceil(gapsize/2)))>0);

 C=imdilate(bwskel(B),strel('sphere',1))|A;
 
 imshow(C)

[KALYAN ACHARJYA]

+1 Nice one sir

[Stelios Fanourakis]

I use exactly the same example but it stucks on running and I needed to pause it. Do you know if this take a lot of time to compute?

[Matt J]

I'm guessing imcrop is wating for you to drag a cropping window across the image.

[Stelios Fanourakis]

I'm not using the imcrop line. I use those lines

 x = imread('1pp.bmp');
>>  gapsize=36;
  
 B=(imdilate(x,strel('sphere',ceil(gapsize/2)))>0);
 C=imdilate(bwskel(B),strel('sphere',1))|x;
 
 imshow(C)

But it stucks and still I have to shut it down

[Stelios Fanourakis]

It just stopped, providing a very nice result. However, the big gap in my curve still exist, although smaller gaps were filled and connected. The bigger the gapsize I will assign, the bigger gaps will be connected?

[Matt J]

Yes.

[Stelios Fanourakis]

Yes it's perfect. Thanks a ton Matt. It is exactly what I needed. Although it takes more than 2 minutes in my case and I don't know the reason it does take so long.

[Matt J]

You could replace the dilation step with fft-based convolution for faster speed. This would make sense if you are using very large gap sizes.

[Image Analyst]

Or use traditional edge linking baed on linking closest endpoints, which should be even faster.

[Stelios Fanourakis]

Can you please give an example Image Analyst?

[Stelios Fanourakis]

Matt how your code can be used in my image linking the first white pixel lines from the bottom of the image or the last white pixel lines (lines that look like dots) as you see the image from top to bottom.

I attach the image

[Matt J]

I don't understand how the linking is supposed to work. I suspect, in any case, that it will require a very different method, since it is a very different kind of image. A new thread may be a good idea.

Answer 3

[Image Analyst]

You might look to the "edge linking" link to the right for alternative methods. The method I like, which may be faster than dilation followed by skeletonization is to call bwmorph(bw, 'endpoints') and then use imline() to draw a line connecting endpoints closer than some distance that you specify.

[Mateusz Matysiak]

Image Analyst,

Please explain the imline part. Please give an example.

Thank you