Centerline and bounding curve in image

I have an Image to which I want to fit a centerline and extract the bounding curves of the object. Here is the image and the best that I could manage with my code.

I = img(501:3500,501:5500);

im_bin = imbinarize(I);

se = strel('disk',3);

opened_im = imopen(im_bin,se);

BWdfill= imfill(opened_im, 4,'holes');

BWoutline= bwperim(BWdfill);

Segout = I;

Segout(BWoutline) = 255;

for indx = 1:5000

temp = find(BWoutline(:,indx)==1);

if isempty(temp)==1

temp= NaN;

else

end

top(indx) = max(temp);

bottom(indx) = min(temp);

end

top_temp = sgolayfilt(top,polsgf, winsgf);

top1 = fillmissing(top_temp,'movmedian',10);

bottom_temp = sgolayfilt(bottom, polsgf, winsgf);

bottom1 = fillmissing(bottom_temp,'movmedian',10);

[maxR, Idx1] = max(I,[],1);

const1 = mean(Idx1,"omitnan");

for k = 1:length(Idx1)

if Idx1(:,k) == 1

Idx1(:,k) = NaN;

else

end

end

Idx1 = fillmissing(Idx1,"constant",const1) ;

Cline = sgolayfilt(Idx1, polsgf, winsgf);

Ignore the white line in the output image that is a scale bar.

Answer 1

[Roy Kadesh]

If this complex code doesn't work, can't you try something easy?

img=imread('1.png');
I = img(501:3500,501:5500);
im_bin = imbinarize(I);

mmm=zeros(size(im_bin,2),3);
for col=1:size(im_bin,2)
    ind=find(im_bin(:,col));
    mmm(col,:)=[min(ind) mean(ind) max(ind)];
end

figure(1),clf(1)
imshow(I)
hold on
plot(mmm)

This works better and should be easier to fine-tune.

[Prabhat Srivastava]

Thank you. I have tried this. I have a number of such images so algorithm fails in some images because they don't have consistent structure. The process that you have suggested is what I am trying too. This certainly is more elegant than mine. Thank you very much.

Answer 2

[Image Analyst]

You can either scan the image columns, or use bwskel. In the attached code I do it both ways.

If you want, you could smooth the rows.

[Prabhat Srivastava]

Thank you. bwskel doesn't work because I have many such images and some of them are really hard to process owing to artefacts. I guess scanning across the columns will be a better idea. The problem comes when I smoothen the centerline signal. It either goes out of the bounds or becomes tangent to the bounds that is what I don't want. Still, many thanks :)

[Image Analyst]

The advantage of the skeleton is that the center will be the center even if the blob is tilted (if that's what you want). The scanning method only looks in the vertical direction (1-D), not 2-D like the skeleton. These approaches could differ at times in where they find the centerline, with the skeleton being the more accurate way. Imagine if your blob had a shape like a C or an S instead of something horizontal. What would the column scanning way give you? Yeah, exactly -- now you get it. A line straight through the half way point vertically (actually even goes outside the blob itself), it will NOT be a centerline following the curve that is the blob. So, beware.

[Prabhat Srivastava]

I see!! Sure, I will experiment with these to see which one gisve the best estimation. Thank you.