How to fill a large hole?

The imfill function doesn't fill the large hole. How can I fill it?

Or, more generally:

How can I separate the image from the background?

Using a very large 'disk' might work to fill the hole, but it would also make the subject too large.

Any suggestions?

Thank you!

I = imread("sample_image.png");
BW = imbinarize(I);
se = strel('disk',40);
closeBW = imclose(BW,se);
BW2 = imfill(closeBW,'holes');
figure
imshowpair(I,BW2,'montage')

Answer 1

[DGM]

I don't have the original image, but I'm going to assume that your image is RGB. Use im2gray() to make sure that the incoming image is single-channel; otherwise, imbinarize() will treat it as a volumetric image, which causes problems further along the line with imfill().

[John]

It’s a gray image.

[DGM]

The given image is RGB -- but only because it's a screenshot. I can only guess the problem because I have to guess at the size and depth of the original.

If the image is RGB, it will be treated as a volumetric image. When imfill() is given the resultant MxNx3 logical image, it will also treat it as a volumetric image. In this assumption, the holes are not filled because they are not closed. They are fully-connected to the image boundary (at the top and bottom pages of the volume).

inpict = imread('image.jpg'); % this is RGB (MxNx3)
BW = imbinarize(inpict);
se = strel('disk',40);
closeBW = imclose(BW,se);
BW2 = imfill(closeBW,'holes');
imshowpair(inpict,BW2,'montage','scaling','none')

If the image is a single channel image, the binary image is as well. In this case, the holes are closed and isolated from the image boundary.

inpict = im2gray(inpict); % now it is I (MxNx1)
BW = imbinarize(inpict);
se = strel('disk',40);
closeBW = imclose(BW,se);
BW2 = imfill(closeBW,'holes');
imshowpair(inpict,BW2,'montage','scaling','none')

If imfill() isn't filling the area, then either:

• the image is not single-channel
• the hole is not fully enclosed
• there is some other function named imfill() which is shadowing IPT imfill()
• something else

There's one other clue I should point out here. Note that I used different arguments in my calls to imshowpair(). There's a reason for that. Let's see what happens if I pass a single-channel image (the RHS) to imshowpair() with the original usage:

imshowpair(inpict,BW2,'montage')

Notice that the contrast is much higher than in any of the previous examples. That's because by default, imshowpair() will normalize grayscale images for display, which effectively stretches the contrast. This display normalization is not applied to RGB images. I used the extra scaling arguments in the two examples above so that they appeared the same.

Given that the image in the screenshot is clearly not normalized, I have to conclude that either your image is not a single-channel image, or the screenshot does not correspond to the specific usage of imshowpair() that is shown.

Answer 2

[Walter Roberson]

The provided JPEG image is RGB, not gray.

The provided JPEG image has a white border around it.

I = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1823171/image.jpeg');
I = rgb2gray(I);
I = imclearborder(I);
BW = imbinarize(I);
se = strel('disk',40);
closeBW = imclose(BW,se);
BW2 = imfill(closeBW,'holes');
imshow(BW2)

●