function [features, count_circles, centers, separation12] = findcircles(input)
% GENERAL PURPOSE: To provide an example of feature-extraction in a binary image.
% This example could be adapted for other specific requirements.
%
% NOTE: Requires IMCIRCLE.M, also provided in the MATLAB image userfiles.
%
% SPECIFIC PURPOSE: To identify features in a binary image, to locate circles among the
% features, their centers and the distance between the first two.
%
% SYNOPSIS: [features, count_circles, centers, separation12] = findcircles(input)
%
% DESCRIPTION: Features are assumed to be made up of ones, against a background of zeros.
% The program makes an effort to remove outliers and stray pixels (line 47).
% Features are isolated using MATLAB's morphological routine BWLABEL.
% Conceptually, a square is drawn around each separate feature, then the
% feature is compared with a circle occupying the same square. If it
% resembles the circle to within a certain tolerance (line 40), the feature
% is counted as a circle. Features are not considered if they are smaller than
% a preset size (line 43). The program also finds the distance between the
% centers of the first two circles found. Circles may be missed if they are
% only partly in the image (i.e. at the edges).
%
% EXAMPLE: Experiment on a blank image by adding features from among the 5 below.
% (Include the file imcircle.m in the MATLAB working directory -- see NOTE above)
%
% myimage = zeros(100); % blank image
% myimage(51:56,52:57) = imcircle(6); % feature 1 (circle)
% myimage(3:29,13:39) = imcircle(27); % feature 2 (circle)
% myimage(51:65,82:96) = imcircle(15); % feature 3 (circle)
% myimage(71:90,12:31) = ones(20); % feature 4 (square)
% myimage(22:36,71:96) = ones(15,26); % feature 5 (rectangle)
% imshow(~myimage),shg % display image
%
% [features,count_circles,centers,separation12] = findcircles(myimage)
%
% features = 5, count_circles = 3; center1 = [15.5 25.5]; center2 = [53 54];
% center3 = [57.5 88.5], separation12 = 47.10;
tolcirc = 0.15; % Tolerance for deciding if a feature is a circle. Range [0,1].
% To relax the specification, increase tolcirc (circles can be more imperfect).
% To tighten the specification, decrease tolcirc (only good circles qualify).
tolsize = 3; % Tolerance for rejecting small features (diameters). Range [1,2,3...]
% To accept all features (diameters), even one pixel, set to 1.
% To reject ever larger features, increase to higher integers.
disp(' '),disp('To operate FINDCIRCLES, you must include IMCIRCLE.M in the working directory.')
disp('IMCIRCLE.M is provided in the image section of MATLAB user-donated files.'),disp(' ')
if length(find(input~=0 & input~=1)) > 0, error('Input must be binary'), end
input = bwmorph(bwmorph(bwmorph(input,'spur'),'clean'),'fill');
[numbers,features] = bwlabel(input);
count_circles = 0; centers = [0 0]; separation12 = [];
for i = 1:features
[I,J] = find(numbers == i);
rows = min(I):max(I); columns = min(J):max(J);
rows = min(rows):min([size(input,1) min(rows)+max([length(rows) length(columns)])-1]);
columns = min(columns):min([size(input,2) min(columns)+max([length(rows) length(columns)])-1]);
testitem = input(rows,columns); % testitem is a square containing the feature.
diameter = max([length(rows) length(columns)]);
circle = imcircle(diameter);
circle = circle(1:length(rows),1:length(columns)); % circle is compared to testitem.
if (length(find(xor(testitem,circle))) < round(tolcirc*diameter^2)) & (diameter >= tolsize)...
& length(rows)/length(columns) >= 0.5 & length(rows)/length(columns) <= 2,
count_circles = count_circles + 1;
center_rows = rows(1):rows(1)+diameter-1;
center_columns = columns(1):columns(1)+diameter-1;
centers(count_circles,:) = mean([center_rows;center_columns]')-0.5;
end
end
if count_circles > 1, separation12 = norm(centers(2,:)-centers(1,:)); end
