function [Dictionary] = DictionaryFormationNV(train_features,Options)

arguments (Input)

train_features {mustBeA(train_features,"cell")}

Options.Centers {mustBeInteger,mustBePositive} = 200

end

% Preallocation of the size of the Centers variable as it will be of

% use in the use of k-means

Dictionary = zeros(Options.Centers,128);

% Due to the variable size of the images, the SIFT descriptors matrix

% is not trivial to be computed in order to preallocate it.

training_SIFT_matrix= [];

tic

% Extract the SIFT descriptor data as they are stored as ArrayDatastore

% variables and create a collective matrix that contains all the

% descriptors of all the training images

for i = 1:length(train_features)

feature_data = read(train_features{i});

training_SIFT_matrix = [training_SIFT_matrix;feature_data.data];

end

% Get the size of the SIFT matrix

[numKeypoints DescDim] = size(training_SIFT_matrix);

% The number of Centers provided by the user must not exceed the total

% number of keypoints as they are formed by the collective matrix of

% SIFT descriptors, e.g. here SIFT_matrix's first dimension.

validateattributes(Options.Centers,'numeric',{'<',numKeypoints})

[bestCentroids, bestCost, timeElapsed] = miniBatchKMeansNV(training_SIFT_matrix);

% Display results

disp('Best cost:')

disp(bestCost);

disp(['Elapsed Time: ' num2str(timeElapsed) ' seconds']);

Dictionary = bestCentroids;