How to have automated thresholding to deal with disparities between images

Hi,

I have huge sets of images of some dark circles against a background, the darkness of the background can vary. What I am trying to do is load the image, threshold it, convert to B&W, then take centroids of the circles and set a crop rectangle to crop the image based on the average of the centroid positions, then I want to save that cropped image into a cell (or other if there's a better method) and then stitch the cropped images together in the sequence they were saved - e.g. image1 on the far left, image2 adjacent to its right.

So far my code is stumbling at the automated thresholding, I tried Otsu's method and received a similar error:

Index exceeds array bounds.
Error in crop_program_08 (line 14)
    pic_testing=(files(k).name)

Code:

close all
clear all
clc

%%Get files
files=dir('/Users/imagexpertinc/Desktop/piccies/*.png');

N=numel(files)
ImageCell = cell(N,1);
%%************************************************************************
image1=imread(files(1).name);
% 
for k=1:2
    pic_testing=(files(k).name)
    image=imread(files(k).name);
%%Show first Image
figure
  image1=image;
  imshow(image1)
  title('1')
  level = graythresh(image1)
  threshold = 70; % custom threshold value
% image1_bw = image1 > threshold;
image1_bw = imbinarize(image1,level)
[centers,radii] = imfindcircles(image1_bw,[8 15],'ObjectPolarity','dark','Sensitivity',0.85)
figure
imshow(image1_bw)
%  h = viscircles(centers,radii);
meanOfcenters=mean(centers(:,1));
MaxRadii=max(radii(:,1));
LimLow=meanOfcenters-(2*MaxRadii);
LimHI=meanOfcenters+(2*MaxRadii);
centersLimits={LimLow, LimHI};
centersLimits=cell2mat(centersLimits);

[HEIGHT,~]=size(image1_bw);
WIDTH=150;
XMIN= meanOfcenters-(WIDTH/2);

YMIN=1;
rect=[XMIN YMIN WIDTH HEIGHT];%where RECT is a 4-element vector with the form [XMIN YMIN WIDTH HEIGHT];

[x,y,I2,rect] = imcrop(image1_bw,rect);
% whos I2;

[CroppedImage] = imcrop(image1,rect);

ImageCell{k}=CroppedImage;

end

AA=ImageCell{:};
ImageCell = cat(2,ImageCell{:}) ;

hFig=figure('units','normalized','outerposition',[0 0 1 1]);
set(0,'CurrentFigure',hFig)
imshow(ImageCell);

Answer 1

[Anton Semechko]

Here is a demo that uses automatic multi-level thresholding to isolate the disks:

im_files={'https://www.mathworks.com/matlabcentral/answers/uploaded_files/122855/picture1.png' ...
          'https://www.mathworks.com/matlabcentral/answers/uploaded_files/122854/picture2.png'};

figure  
for i=1:2

      % Get the image
      im=imread(im_files{i});

      % Partion image into 4 classes
      [~,~,LUT,~]=FastFCMeans(im,4);
      L=LUT2label(im,LUT);

      % Isolate disks
      bw=L==1;
      bw=bwareaopen(bw,10,8); % remove 8-connected blobs (if any) with fewer than 10 pixels

      % Visualize
      subtightplot(2,2,2*i-1), imshow(im)
      subtightplot(2,2,2*i),   imshow(bw)

end

Segmentation functions used in this demo can be downloaded from here

'subtightplot' function can be found here

[Stephen Devlin]

Hi Anton, I am unable to replicate your result, in mine I see no white in the b&w image, have attached a screenshot. Best regards,

Steve

[Anton Semechko]

Do you have any other copies of 'FastFCMeans' function that you may have downloaded from here? If so, delete it, and use the newer version from here.