Particle recognition from image with regionprops

Hi,

I'm trying to write a script for automated particle recognition and subsequent size analysis from SEM images (scanning electron microscopy). I tried to adapt a published script ("A simple algorithm for measuring particle size distributions on an uneven background from TEM images", Gontard et al., 2011) to my needs:

% Selecting an image
[file,ruta]=uigetfile({'*.tif';'*.bmp'},'Select a SEM image');
cd(ruta);
fc_1 = imread(file(:,:,1));
fc_2 = imcomplement(fc_1);
fc = imcrop(fc_2,[1 1 1024 600]);
imshow(fc);

% Selecting subimages
[N,M] = size(fc);
ns=input('Number of divisions = '); min_nm=input('Minimum size (nm) = ');
mag = input('Magnification (x) = ');
scale = mag^(-1.002)*936937; % nm per pixel
min_px = round(min_nm / scale);
x = fix(N/ns);
y = fix(M/ns);
for sx=1:x:N-x
    for sy=1:y:M-y
        sp=fc(sx:sx+x-1,sy:sy+y-1);

          % Thresholding
          T = graythresh(sp);
          spT = im2bw(sp,T);
          g(sx:sx+x-1,sy:sy+y-1)=spT;
      end
  end
  g2 = imopen(imcomplement(g),strel('disk',min_px));
  [labeled,a]=bwlabel(g2,4);
  points = regionprops(labeled,'Centroid','PixelList');
  [B,L2,N2] = bwboundaries(labeled,4,'noholes');

% Draw segmented particles
imshow(fc);
hold on
for s=1:numel(points)
    boundary = B{s};
    if(s > a)
        plot(boundary(:,2), boundary(:,1), 'g','LineWidth',1);
    else
        plot(boundary(:,2), boundary(:,1), 'r','LineWidth',1);
    end
end
hold off

% Histogram
graindata = regionprops(labeled,'basic');
areap = [graindata.Area];
t=0;
for s=1:numel(points)
    t = t + 1;
    sizes_px(t) = 2*sqrt(areap(s)/pi);
end
figure;
nm = mag^(-1.002)*936937;
sizes_nm = nm*sizes_px;
hist(sizes_nm,30);
Mean=mean(sizes_nm), Standard_deviation=std(sizes_nm)

However, it's not working as expected yet. To give you an idea of how the pictures might look like, here's an example:

Now my problem is, that some particles are not at all recognized. In other cases, two particles that overlap are considered as one big (next pictures shows the results from the script).

I know that this is a complex task, but it might already be helpful to set an area limit, like the opposit of the minimum area. But I don't know if that's possible. Any ideas?

Thanks so much!

Answer 1

[Image Analyst]

I'd try imfindcircle() and see how that does. You could also experiment with imtophat() and imbottomhat().

Also, you should read this: http://www.mathworks.com/matlabcentral/answers/13205-tutorial-how-to-format-your-question-with-markup

[Roman]

Thanks, I was already wondering why my question was so short, but I didn't realize it was because of the code...

I will have a look at the commands, sounds promising. Thanks again!

[Image Analyst]

Here's a demo for you: http://www.mathworks.com/matlabcentral/fileexchange/34365-findcirclesgui

Answer 2

[Sergey]

I'd recommend you imtophat/imbothat to make background more regular and than marker-guided watershed (combination of imregionalmax, imimposemin and watershed: you impose minima the background detected according to gray level, I usually choose <0.5*graythresh(), and regional maxima on complementary - objects are darker - images, and then perform the watershed). The approach worked for pollen TEM images.