# how can i remove the white object from binary image

14 views (last 30 days)
R G on 13 Apr 2018
Edited: Image Analyst on 18 Apr 2018
I have a image with my needed object (green) and the noise (white). I want to delete the noise but have no idea for that. anyone can help

Gopichandh Danala on 13 Apr 2018
You can use regionprops to compute circularity, eccentricity, or use major axis and minor axis lengths to find required shape of blob.
I used major and minor axis:
dilate_img = imdilate(img,strel('disk',2));
BW = bwlabel(dilate_img);
blobprops = regionprops(BW, 'MajorAxisLength','MinorAxisLength', 'Eccentricity');
allMajorAxisLength = [blobprops.MajorAxisLength];
allMinorAxisLength = [blobprops.MinorAxisLength];
allEccentricity = [blobprops.Eccentricity];
MajorbyMinor = allMajorAxisLength./allMinorAxisLength;
[B,I] = sort(MajorbyMinor);
BW2 = ismember(BW,I(1));
% Using eccentricity
% [B,I] = sort(allEccentricity); % eccentricity = 0: circle, = 1: line
% BW2 = ismember(BW,I(1));
figure,
subplot(131), imshow(img), title('Original');
subplot(132), imshow(BW), title('Label');
subplot(133), imshow(BW2), title('desired');

R G on 14 Apr 2018
what is the meaning of imdilate step
dilate_img = imdilate(img,strel('disk',2));
I mean why you need to make the line more fat since the object is fat and short, the noise is thin and long. Can we use that characteristic as an advantage.
and this line is kinda tricky.
BW2 = ismember(BW,I(1));
Since every object have a different "MajorbyMinor " value and I don't know how many object that I need. That code make us always can get only one object. I think a threshold of "MajorbyMinor " or "eccentricity" will work better.
that is my first idea for this problem but I don't know how to make it possible in code. I am not quite sure about that, can you explain more about your idea.
Gopichandh Danala on 17 Apr 2018
The answer is always based on your question. You specifically asked for the green object in your image, so I answered accordingly. For imdilate, ismember, check the documentation. The reason to use dilate in this case is to connect the objects on the right to make it as one. so, it's easy to identify it as not being a round object. To make this more generic to identify multiple regions which satisfy a certain threshold condition.
MajorbyMinor = allMajorAxisLength./allMinorAxisLength;
thresh_value = 4; % set a threshold value
indxs = MajorbyMinor < thresh_value;
% [B,I] = sort(MajorbyMinor);
BW2 = ismember(BW,find(indxs));
It all depends on your requirement. i.e. the shape, size of the object. Adjust the threshold value to meet your requirement.

Image Analyst on 15 Apr 2018
You didn’t say what was unique about the object but it looks like it might be the size of the bounding box. So why don’t you compute that with regionprops()?

R G on 15 Apr 2018
my bad, I am kinda amateur about this. I think the object is short and fat. the noise is thin and long. So it may concern to major axis and minor axis. That is all I can say.
how do you see it concern with the bounding box. Can you explain more detail?
Image Analyst on 18 Apr 2018
Compute the bounding box.
labeledImage = bwlabel(binaryImage);
props = regionprops(labeledImage, 'BoundingBox');
Then decide which ones to keep, like if the width to height ratio is between 0.5 and 3, or whatever.
indexesToKeep = false(1, length(props));
for k = 1 : length(props)
width = props(k).BoundingBox(3);
height = props(k).BoundingBox(4);
if width/height > 0.5 || width/height < 3
% It's a fairly square box, so keep it.
indexesToKeep(k) = true;
end
end
% Extract just acceptable blobs
indexesToKeep = find(indexesToKeep);
binaryImage2 = ismember(labeledImage, indexesToKeep);
imshow(binaryImage2);