def __init__(self, dir_name, verbose=False):
"""Load a collection of images from a directory"""
self.image_coll = ImageCollection.from_directory(dir_name,
verbose=verbose)
self.num_images = self.image_coll.num_images
iflat = self.image_coll.flat()
data_mean = iflat.mean(axis=1)
img_flat_norm = iflat - data_mean[:, N.newaxis]
# Compute singular values and U matrix
umat, sigma, vtmat = N.linalg.svd(img_flat_norm, 0)
# Array of 'eigenimages' in normal (not flattened) format
im_shape = self.image_coll.images[0].shape
eig_img = N.empty((self.num_images, im_shape[0], im_shape[1]),
umat.dtype)
for i in range(self.num_images):
eig_img[i] = N.reshape(umat[:, i], im_shape)
# Store in object all these
self.sigma = sigma
self.umat = umat
self.vtmat = vtmat
self.im_shape = im_shape
self.eig_img = ImageCollection(eig_img)
self.img_flat_norm = img_flat_norm
self.data_mean = data_mean
# Truncation index, by default we don't truncate
self.truncate_idx = self.num_images - 1
def __init__(self,dir_name,verbose=False):
"""Load a collection of images from a directory"""
self.image_coll = ImageCollection.from_directory(dir_name,
verbose=verbose)
self.num_images = self.image_coll.num_images
iflat = self.image_coll.flat()
data_mean = iflat.mean(axis=1)
img_flat_norm = iflat - data_mean[:,N.newaxis]
# Compute singular values and U matrix
umat,sigma,vtmat = N.linalg.svd(img_flat_norm,0)
# Array of 'eigenimages' in normal (not flattened) format
im_shape = self.image_coll.images[0].shape
eig_img = N.empty((self.num_images,im_shape[0],im_shape[1]),umat.dtype)
for i in range(self.num_images):
eig_img[i] = N.reshape(umat[:,i],im_shape)
# Store in object all these
self.sigma = sigma
self.umat = umat
self.vtmat = vtmat
self.im_shape = im_shape
self.eig_img = ImageCollection(eig_img)
self.img_flat_norm = img_flat_norm
self.data_mean = data_mean
# Truncation index, by default we don't truncate
self.truncate_idx = self.num_images-1
# Add the weighted eigenfaces to form the face
eig_composition += test_coeffs[i] * eig_img[i]
# Add back the data mean (subtracted before composition)
eig_composition += data_mean.reshape(im_shape)
P.imshow(eig_composition, P.cm.gray)
else:
P.imshow(face, P.cm.gray)
P.title('Original image')
# 'main' script
# Load the training images from directory 'data_train/'
imtrain = ImageCollection.from_directory('data_train/')
# Compute the data mean and normalise
data_mean = imtrain.flat().mean(axis=1)
img_flat_norm = imtrain.flat() - data_mean[:, N.newaxis]
# Compute singular values and U matrix
umat, sigma, vtmat = N.linalg.svd(img_flat_norm, 0)
# Create an array of 'eigenfaces' in normal (not flattened) format
im_shape = imtrain.im_shape
eig_img = N.empty((imtrain.num_images, im_shape[0], im_shape[1]), umat.dtype)
for i in range(imtrain.num_images):
eig_img[i] = N.reshape(umat[:, i], im_shape)
# Add the weighted eigenfaces to form the face
eig_composition += test_coeffs[i]*eig_img[i]
# Add back the data mean (subtracted before composition)
eig_composition += data_mean.reshape(imshape)
P.imshow(eig_composition, P.cm.gray)
else:
P.imshow(face, P.cm.gray)
P.title('Original image')
# 'main' script
# 'main' script
# Load the training images from directory 'data_train/'
imtrain = ImageCollection.from_directory('data_train/')
# Compute the data mean and normalise
data_mean = imtrain.flat().mean(axis=1)
img_flat_norm = imtrain.flat() - data_mean[:,N.newaxis]
# Compute singular values and U matrix
umat,sigma,vtmat = N.linalg.svd(img_flat_norm,0)
# Create an array of 'eigenfaces' in normal (not flattened) format
im_shape = imtrain.im_shape
eig_img = N.empty((imtrain.num_images,imshape[0],imshape[1]),umat.dtype)
for i in range(imtrain.num_images):
eig_img[i] = N.reshape(umat[:,i],imshape)