CODEWITHSUNDEEP
matlabimage-processingimage-morphology
Tapio
Tapio

Reputation: 1642

Morphological thickening/dilation while trying to preserve shape in MATLAB

I'm thickening a binary image. I wished to grow the label n pixels to each direction.

Original image:

Original label image

At first I used the function bwmorph(I, 'thicken', 25) and got this image :

After morphological thickening

Which is no good. The thickening seems to use the structuring element [0,1,0;1,1,1;0,1,0], so it always transforms circles into diamonds. imdilate with the mentioned structuring element results in the same output.

Next I tried dilating the original image iteratively for n=25 times with the structuring element[1,1,1; 1,1,1; 1,1,1] and got the following image:

iterative dilation

The original shape is now completely gone.

I understand that dilation will always distort the border to some degree. I experimented with the structuring element 'disk' (r=5, dilated 5 times) and got fairly good results:

enter image description here

Is this as good as it gets? Which structuring element should I choose if I wish to preserve the original circular shape as well as possible? Are there any good rules of thumbs for finding the right size for a particular dilation distance (which can vary from 10-100 and the labels vary from circular to elliptical)? Is there a better way of growing a binary image in all the directions while trying to keep the original shape?

Is there a way to keep the Euler characteristic -preserving quality of thicken while changing the structuring element to something more suitable?

Upvotes: 2

Views: 1280

Answers (2)

Erfan
Erfan

Reputation: 1927

Try this:

bw = imread('bw.png');
figure; imshow(imdilate(bw, strel('arbitrary', bw)))

I am using the image itself for extending it. It gives a more shape-preserving growth:

UPDATE: As @Tapio said, one can easily adjust the resize factor using imresize:

 figure; imshow(imdilate(bw, strel('arbitrary', imresize(bw, 0.5))))

with the following output:

enter image description here

Upvotes: 3

FiReTiTi
FiReTiTi

Reputation: 5898

If you use only unitary structuring elements, then you will always face the problems you described. If you want to dilate a shape with a structuring element of size N, do not use N times a structuring element of size 1, else you will end up with beautiful unexpected deformation.

If you want to do a dilation of size N, try a disk of radius N. However, it will be much longer to process.

An other solution is to use a distance map: the shape is the seed (distance 0), then everything at a distance smaller than N is your result. It will be pretty similar to the dilation of size N.

Upvotes: 1

Related Questions