def show_reconstruction(n_training_img=5):
"""
Show the images corresponding to the reconstruction process, namely:
- Original face
- Mean image
- Eigenfaces
- Reconstructed face
"""
train_img, train_labels, test_img, test_labels = load_images(
n_training_img)
nearest_neighbor = NN()
nearest_neighbor.train(train_img, train_labels, 10)
face_idx = np.random.randint(0, len(train_labels))
plt.imshow(shape_image(train_img[face_idx]), cmap="gray")
plt.title("Original face")
plt.axis('off')
plt.figure()
eigenfaces = np.zeros(nearest_neighbor.eigenfaces.shape)
for i in range(nearest_neighbor.eigenfaces.shape[1]):
eigenfaces[:, i] = nearest_neighbor.face_space_coord_train[
face_idx] * nearest_neighbor.eigenfaces[:, i]
plt.imshow(shape_image(nearest_neighbor.mean_face), cmap="gray")
plt.title("Mean face")
plt.axis('off')
plt.figure()
for i, eigenface in enumerate(eigenfaces):
plt.subplot(1, eigenfaces.shape[0], i + 1)
plt.imshow(shape_image(eigenface), cmap="gray")
plt.axis('off')
plt.subplots_adjust(wspace=0, hspace=0)
plt.suptitle("Eigenfaces")
plt.figure()
plt.imshow(shape_image(
nearest_neighbor.get_reconstructed_faces(train_img)[face_idx]),
cmap='gray')
plt.title("Reconstructed image")
plt.axis('off')
plt.show()
def show_reconstructed(n_training_img, k_ppal_components, n_images):
"""
Show side by side the original image and the reconstruction using different number of principal components (eigenfaces)
:param n_training_img: Number of training images used per person
:param k_ppal_components: Array with the number of eigenfaces to use
:param n_images: Number of images of people to show
"""
train_img, train_labels, test_img, test_labels = load_images(
n_training_img)
# Randomize the selection of faces to show
face_ids = [i for i in range(len(train_labels))]
np.random.shuffle(face_ids)
# Reconstruct the images from the training data
reconstructed = []
for n_k in k_ppal_components:
nearest_neighbor = NN()
nearest_neighbor.train(train_img, train_labels, n_k)
reconstructed.append(
nearest_neighbor.get_reconstructed_faces(train_img))
# Plot the results
fig, axarr = plt.subplots(n_images, 1 + len(k_ppal_components))
axarr[0, 0].set_title("Original")
for j, n_k in enumerate(k_ppal_components):
axarr[0, j + 1].set_title("K = " + str(n_k))
for i in range(n_images):
face_id = face_ids[i]
axarr[i, 0].imshow(shape_image(train_img[face_id]), cmap="gray")
axarr[i, 0].axis('off')
for j, n_k in enumerate(k_ppal_components):
axarr[i, j + 1].imshow(shape_image(reconstructed[j][face_id]),
cmap="gray")
axarr[i, j + 1].axis('off')
plt.subplots_adjust(wspace=0, hspace=0)
fig.suptitle(f'Reconstruction analysis. {n_training_img} training images')
plt.show()
