Skip to content

Instantly share code, notes, and snippets.

@abrahamhurtado

abrahamhurtado/Create_Features.m

Last active March 30, 2019 04:58
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save abrahamhurtado/795695bf40cf8f451c6ca7696e5f1988 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save abrahamhurtado/795695bf40cf8f451c6ca7696e5f1988 to your computer and use it in GitHub Desktop.
Download ZIP
Segmentacion
Raw
Create_Features.m
function [ Features ] = Create_Features(imagen, num_rows, num_cols)
total_elems = num_rows * num_cols;
Features = zeros(total_elems, 5);
x = 1;
y = 1;
for i=1:total_elems
Features(i,1) = x;
Features(i,2) = y;
Features(i,3) = imagen(x,y,1);
Features(i,4) = imagen(x,y,2);
Features(i,5) = imagen(x,y,3);
x = x + 1;
if (mod(i, num_rows) == 0)
x = 1;
y = y + 1;
end
end
end
Raw
Denormalize_Centroids.m
function [ denormalized ] = Denormalize_Centroids(Centroids, Features)
denormalized = zeros(size(Centroids));
for i=1:5
valor_max = max(Features(:,i));
valor_min = min(Features(:,i));
denormalized(:, i) = round((Centroids(:,i) * (valor_max - valor_min)) + valor_min);
end
end
Raw
init_centroids.m
function [ centroids ] = init_centroids( Features_Norm, K )
centroids = zeros(K, size(Features_Norm, 2));
index_registros_representativos = randperm(length(Features_Norm), K);
for i=1:K
centroids(i,:) = Features_Norm(index_registros_representativos(i), :);
end
end
Raw
K_Means.m
function [ dataset, predictions, centroids, centroids2, iterations ] = K_Means(image_og, K, w)
[num_rows, num_cols, dimensions] = size(image_og);
num_elems = num_rows * num_cols;
image = zeros(num_rows, num_cols, dimensions);
image(:,:,1) = image_og(:,:,1);
image(:,:,2) = image_og(:,:,2);
image(:,:,3) = image_og(:,:,3);
% image_new = zeros(num_rows, num_cols, dimensions);
Features = Create_Features(image, num_rows, num_cols);
Features_Norm = normalize_matrix(Features, w);
centroids = init_centroids(Features_Norm, K);
centroids2 = zeros(size(centroids));
predictions = zeros(num_elems, 1);
iterations = 0;
% El algoritmo se ejecutar� hasta que centroids y centroids2 sean iguales,
% es decir, que converjan.
while isequal(centroids, centroids2) == false
iterations = iterations + 1;
% Recorremos por el total de datos, en este caso 633
for i = 1:num_elems
% Obtenemos la distancia euclidiana del registro i del dataset
% respecto a todos los centroides.
% bsxfun(@minus, centroides, dataset(i, :)) nos permite a cada
% rengl�n de la matriz de centroids, restarle el registro ubicado
% en dataset(i), una operaci�n que no se puede ejecutar
% naturalmente y que hubiera requerido otro ciclo for.
% sum(@minus, centroides, dataset(i, :))', se hace la suma de la
% matriz transpuesta porque las sumas de matrices se hacen columna
% por columna, como est� acomodada originalmente la matriz
% centroids 8 x 18, la suma resultar�a en una matriz horizontal de
% 1 x 18, nosotros queremos 1 x 8.
[~, claseAsignada] = min(sum(bsxfun(@minus, centroids, Features_Norm(i,:))'.^2));
predictions(i,:) = claseAsignada;
end
% centroids2 ahora almacena centroids
centroids2 = centroids;
% Recorremos por el total de clases que tenemos, en este caso 8
for j = 1:K
% Filtramos los registros de dataset seg�n las predicciones para la
% clase j.
registrosAsignadosEnJ = Features_Norm(predictions == j, :);
% El nuevo centroids es la media de las predicciones.
centroids(j,:) = mean(registrosAsignadosEnJ);
end
end
dataset = Features;
end
Raw
main.m
imagen = imread('Tiger1.tiff');
[ dataset, predictions, centroids, centroids2, iterations ] = K_Means(imagen, 3, 1);
denormalized = Denormalize_Centroids(centroids, dataset);
[num_rows, num_cols, dimensions] = (size(imagen));
new_image = zeros(num_rows, num_cols, dimensions);
x = 1;
y = 1;
for i=1:length(predictions)
R = denormalized(predictions(i), 3);
G = denormalized(predictions(i), 4);
B = denormalized(predictions(i), 5);
new_image(x,y,1) = R;
new_image(x,y,2) = G;
new_image(x,y,3) = B;
x = x + 1;
if (mod(i, num_rows) == 0)
x = 1;
y = y + 1;
end
end
NI = uint8(new_image);
figure(1);
imshow(NI);
Raw
normalize_matrix.m
function [ Feature_Norm ] = normalize_matrix( Features, w )
Feature_Norm = zeros(size(Features));
for i=1:5
valor_max = max(Features(:,i));
valor_min = min(Features(:,i));
Feature_Norm(:, i) = (Features(:,i) - valor_min)/(valor_max - valor_min);
end
Feature_Norm(:,1) = Feature_Norm(:,1) * w;
Feature_Norm(:,2) = Feature_Norm(:,2) * w;
end
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment