How to classification use the KNN method with GLCM Ektraction?

How to calculate energy, contrast homogen and correlation used GLCM ektraction and classification use the KNN method?

Answer 1

[Gayathri]

Hi @dinda ayusma,

To calculate energy, contrast, homogeneity and correlation we can take the help of "graycomatrix" and "graycoprops" functions in MATLAB. Now the features extracted using these functions could be given as input to the "fitcknn" function to train a k-nearest neighbour classification. Please refer to the code below to achieve the same.

% Replace with your dataset path 
dataPath = "folder_path"; % Update with your dataset path
imds = imageDatastore(dataPath, ...
    'IncludeSubfolders', true, ...
    'LabelSource', 'foldernames', ...
    'FileExtensions', {'.png', '.jpg'});

% Split dataset into training (70%) and testing (30%)
[imdsTrain, imdsTest] = splitEachLabel(imds, 0.7, 'randomized');
disp(['Number of training images: ', num2str(numel(imdsTrain.Files))]);
disp(['Number of testing images: ', num2str(numel(imdsTest.Files))]);


% Function to extract GLCM features (defined below)
trainFeatures = extractGLCMFeatures(imdsTrain);
testFeatures = extractGLCMFeatures(imdsTest);

% Get labels
trainLabels = imdsTrain.Labels;
testLabels = imdsTest.Labels;


% Create KNN model (k=5 neighbors)
knnModel = fitcknn(trainFeatures, trainLabels, ...
    'NumNeighbors', 5, ...
    'Distance', 'euclidean', ...
    'Standardize', true); % Standardize features for better performance

% Predict on test set
predictedLabels = predict(knnModel, testFeatures);

% Compute accuracy
accuracy = mean(predictedLabels == testLabels);
fprintf('Test accuracy: %.2f%%\n', accuracy * 100);



% Function to extract GLCM features from an imageDatastore
function features = extractGLCMFeatures(imds)
    numImages = numel(imds.Files);
    features = zeros(numImages, 4); % 4 features: contrast, correlation, energy, homogeneity
    
    for i = 1:numImages
        img = readimage(imds, i);
        % Extract GLCM features for the image
        features(i, :) = computeGLCMFeatures(img);
    end
end

% Function to compute GLCM features for a single image
function glcmFeatures = computeGLCMFeatures(img)
    % Convert to grayscale if RGB
    if size(img, 3) == 3
        img = rgb2gray(img);
    end
    
    % Define offsets: [distance angle] for 0°, 45°, 90°, 135°
    offsets = [0 1; -1 1; -1 0; -1 -1]; 
    
    % Compute GLCM
    glcm = graycomatrix(img, 'Offset', offsets,'Symmetric', true);
    
    % Compute GLCM features
    stats = graycoprops(glcm, {'Contrast', 'Correlation', 'Energy', 'Homogeneity'});
    
    
    glcmFeatures = [
        mean(stats.Contrast) ...
        mean(stats.Correlation) ...
        mean(stats.Energy) ...
        mean(stats.Homogeneity)
    ];
end

For more information on different functions used in the code above, please refer to the documentation links given below:

https://www.mathworks.com/help/stats/fitcknn.html

https://www.mathworks.com/help/images/ref/graycomatrix.html

https://www.mathworks.com/help/images/ref/graycoprops.html