url = <\/span>\"https:\/\/www.mathworks.com\/matlabcentral\/answers\/uploaded_files\/880740\/image.jpeg\"<\/span>;<\/span><\/span><\/div><\/div>A = imread(url);<\/span><\/span><\/div><\/div>imshow(A)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>Image credit: Aktham Shoukry. Used with permission.<\/span><\/div>Here is an example of what Aktham meant by \"channel.\"<\/span><\/div><\/img><\/div>This post shows one way to accomplish the task, using erosion, morphological reconstruction, and some image arithmetic.<\/span><\/div>The channels are relatively thin in one dimension, either horizontally or vertically. So, a morphological erosion can be constructed that will keep a portion of each grain, while completely eliminating the channels.<\/span><\/div>First, a bit of preprocessing.<\/span><\/div>Ag = im2gray(A);<\/span><\/span><\/div><\/div>B = imcomplement(Ag);<\/span><\/span><\/div><\/div>imshow(B)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>The channels are about 10 pixels thick, while the square-ish \"grains\" are about 50-by-50. Let's erode with a square structuring element with a size that's in between.<\/span><\/div>C = imerode(B,ones(25,25));<\/span><\/span><\/div><\/div>imshow(C)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>Next, use morphological reconstruction to restore the original grain shapes.<\/span><\/div>D = imreconstruct(C,B);<\/span><\/span><\/div><\/div>imshow(D)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>Complement the image to make the grains dark again.<\/span><\/div>E = imcomplement(D);<\/span><\/span><\/div><\/div>imshow(E)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>Subtracting the original grayscale image from <\/span>E<\/span> gets us very close.<\/span><\/div>F = E - Ag;<\/span><\/span><\/div><\/div>imshow(F)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>G = imbinarize(F);<\/span><\/span><\/div><\/div>imshow(G)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>A final cleanup of the small objects, using either <\/span>bwareafilt<\/span> or <\/span>bwareaopen<\/span>, gives us our result.<\/span><\/div>H = bwareaopen(G,50);<\/span><\/span><\/div><\/div>imshow(H)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>The function <\/span>imoverlay<\/span> gives us a nice visualization of the result. (You could also use <\/span>imfuse<\/span>.)<\/span><\/div>K = imoverlay(Ag,H,[.8500 .3250 .0980]);<\/span><\/span><\/div><\/div>imshow(K)<\/span><\/span><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div><\/div>Note, however, that our result isn't quite perfect. There are partially cropped-off grains along the top border that have been identified as channels. Normally, I would recommend using <\/span>imclearborder<\/span> as a preprocessing step, but in this case, <\/span>imclearborder<\/span> eliminates too many of the desired channels.<\/span><\/div>I am open to suggestions about how to refine this method. Leave a comment and let me know what you think.<\/span><\/div>\r\n