kmeans: add support for grayscale images (results in multi-levels threshold)

Author: dlegland

src/@Image/interp.m

@@ -45,7 +45,7 @@
 %% Compute interpolation
 
 nv = size(point, 1);
-nc = channelNumber(obj);
+nc = channelCount(obj);
 val = zeros(nv, nc);
 
 if nd == 2

src/@Image/kmeans.m

@@ -2,15 +2,17 @@
 % Apply k-means clustering to a multi-channel image.
 %
 %   CLASSES = kmeans(IMG, K)
-%   Compute kmeans clustering of the multi-channel image IMG, using K
-%   classes.
+%   Compute kmeans clustering of the image IMG, using K classes. The result
+%   is a label image with values ranging from 1 no K.
 %
 %   [CLASSES, CENTERS] = kmeans(IMG, K);
 %   Also returns the coordinates of class centers. CENTERS is a K-by-NC
 %   array, where NC is the number of channels of original image IMG.
 %
+%   Requires the statistics toolbox.
 %
 %   Example
+%     % k-means segmentation of a RGB color image
 %     img = Image.read('peppers.png');
 %     cls = kmeans(img, 6);
 %     % display the classes with arbitrary colors
@@ -22,6 +24,16 @@
 %     rgb = label2rgb(cls, map);
 %     figure; show(rgb);
 %
+%     % multi-level segmentation of a grayscale image with k-means
+%     nc = 5;
+%     img = Image.read('cameraman.tif');
+%     [classes, values] = kmeans(img, nc);
+%     res = Image.zeros(size(img),'uint8');
+%     for i = 1:nc
+%         res(classes == i) = values(i);
+%     end
+%     figure; show(res);
+%
 %   See also
 %     label2rgb, fold
 %
@@ -33,26 +45,20 @@
 % Created: 2020-01-15,    using Matlab 9.7.0.1247435 (R2019b) Update 2
 % Copyright 2020 INRAE.
 
-if channelNumber(obj) < 2
-    error('Requires a multi-channel image');
-end
-
 % convert image content into nr-by-nc array
-nr = elementNumber(obj);
-nc = channelNumber(obj);
+nr = elementCount(obj);
+nc = channelCount(obj);
 data = double(reshape(obj.Data, [nr nc]));
 
 % run the kmeans algorithm
+% (use statistics toolbox function)
 [idx, centers] = kmeans(data, k);
 
 % convert to label image, by avoiding zero label
 idx = reshape(idx, size(obj));
 
 % create parent image
-newName = 'classes';
-if ~isempty(obj.Name)
-    newName = sprintf('%s-kmeans%d', obj.Name, k);
-end
-res = Image('Data', idx, 'Type', 'label', ...
-    'parent', obj, 'Name', newName, ...
-    'ChannelNames', {'Class'});
+newName = createNewName(obj, sprintf('%%s-kmeans%d', k));
+res = Image('Data', idx, 'Parent', obj, ...
+    'Type', 'label', 'Name', newName, ...
+    'ChannelNames', {'Class'});

tests/image/test_kmeans.m

@@ -0,0 +1,29 @@
+function tests = test_kmeans
+% Test suite for the file kmeans.
+%
+%   Test suite for the file kmeans
+%
+%   Example
+%   test_kmeans
+%
+%   See also
+%     kmeans
+
+% ------
+% Author: David Legland
+% e-mail: david.legland@inrae.fr
+% Created: 2021-01-27,    using Matlab 9.8.0.1323502 (R2020a)
+% Copyright 2021 INRAE - BIA-BIBS.
+
+tests = functiontests(localfunctions);
+
+function test_Simple(testCase) %#ok<*DEFNU>
+% Test call of function without argument.
+
+img = Image.read('peppers.png');
+
+classes = kmeans(img, 6);
+
+assertTrue(testCase, isa(classes, 'Image'));
+assertTrue(testCase, isLabelImage(classes));
+assertEqual(testCase, max(classes), 6);