Store image into array for classification training (SOM)

Hi there, i currently have over 7000 images that i want to train for classifcation of certain characters. I tried loading them into a matrix but it ends up as a 3D array like this:

arrayImage 128x128x7112 double

clear,clc;
files = dir('*.png');
num = length(files);
arrayImage = zeros(128,128,num);
i = 1;

for file = files'
    img = imread(file.name);
    arrayImage(:,:,i) = img;
    i = i + 1;
end

I think matlab's neural net clustering only reads in a 2D matrix. How can i store them as a 2D array so i can load them into SOM for training. I attached some examples of images i want to store and use for training.

Answer 1

[Umeshraja]

Hi @Justin

I understand you're preparing a large dataset of images for training a Self-Organizing Map (SOM) and need to convert your array of images into a 2D matrix.

Neural network clustering algorithms typically require input in a 2D array format (M x N), where M is the number of observations and N is the number of features. In your case, M would be 7112 (the number of images) and N would be 16384 (the result of reshaping each 128x128 image into a 1D array).

Here's how you can modify your code:

files = dir('*.png');
num = length(files);

% Initialize the 2D matrix
% Each row will represent one image, each column a pixel
arrayImage2D = zeros(num, 128*128);

for i = 1:num
    % Read the image
    img = imread(files(i).name);

    % Ensure the image is in double format
    img = im2double(img);
 
    % Reshape the 2D image into a 1D vector and store it as a row in arrayImage2D
    arrayImage2D(i, :) = reshape(img, 1, []);
end

% Now arrayImage2D is a 2D matrix where each row represents an image
% The size should be [7112, 16384] for your dataset
size(arrayImage2D)

This 2D format should be compatible with MATLAB's Neural Network Toolbox for training a Self-Organizing Map.

For more information on the reshape function, you can refer to the MATLAB documentation:

https://www.mathworks.com/help/matlab/ref/double.reshape.html