def identify(self,test_img):
"""Find the best match from the test set to the provided image."""
norm_test = test_img.flat - self.data_mean
test_coeffs = N.dot(self.utt,norm_test)
diff2 = ((self.proj_c - test_coeffs)**2).sum(axis=1)
minidx = diff2.argmin()
best_err = diff2[minidx]
imshow2(test_img,self.image_coll.images[minidx],
labels = ('Test Image','Best Match: %d' % minidx))
P.title('L2 error: %.2e' % best_err)
def identify(self, test_img):
"""Find the best match from the test set to the provided image."""
norm_test = test_img.flat - self.data_mean
test_coeffs = N.dot(self.utt, norm_test)
diff2 = ((self.proj_c - test_coeffs)**2).sum(axis=1)
minidx = diff2.argmin()
best_err = diff2[minidx]
imshow2(test_img,
self.image_coll.images[minidx],
labels=('Test Image', 'Best Match: %d' % minidx))
P.title('L2 error: %.2e' % best_err)
def verify(test_img, key):
"""Verify visually that a provided image corresponds to a specified image
in the training set.
Two images, the provided test image and the training image 'key', are
displayed side-by-side, along with the l2-norm error. Based on this
information, the user can visually verify that they are the same."""
ref_coeffs = proj_c[key]
norm_test = test_img.flat - data_mean
test_coeffs = N.dot(utt, norm_test)
l2_err = N.linalg.norm(test_coeffs - ref_coeffs, 2)
imshow2(imtrain[key], test_img, labels=('Reference Image', 'Test Image'))
P.title('L2 coefficient error: %.2e' % l2_err)
def verify(test_img,key):
"""Verify visually that a provided image corresponds to a specified image
in the training set.
Two images, the provided test image and the training image 'key', are
displayed side-by-side, along with the l2-norm error. Based on this
information, the user can visually verify that they are the same."""
ref_coeffs = proj_c[key]
norm_test = test_img.flat - data_mean
test_coeffs = N.dot(utt,norm_test)
l2_err = N.linalg.norm(test_coeffs-ref_coeffs,2)
imshow2(imtrain[key],test_img,
labels = ('Reference Image','Test Image'))
P.title('L2 coefficient error: %.2e' % l2_err)
def identify(test_img):
"""Try to find a match for test_img from the training set."""
# Normalise data, compute projections into eigenspace
norm_test = test_img.flat - data_mean
test_coeffs = N.dot(utt,norm_test)
# Find closest match
diff2 = ((proj_c - test_coeffs)**2).sum(axis=1)
minidx = diff2.argmin()
best_err = diff2[minidx]
# Display the matching face
imshow2(test_img,imtrain.images[minidx],
labels = ('Test Image','Best Match: %d' % minidx))
P.title('L2 coefficient error: %.2e' % best_err)
def identify(test_img):
"""Try to find a match for test_img from the training set."""
# Normalise data, compute projections into eigenspace
norm_test = test_img.flat - data_mean
test_coeffs = N.dot(utt, norm_test)
# Find closest match
diff2 = ((proj_c - test_coeffs)**2).sum(axis=1)
minidx = diff2.argmin()
best_err = diff2[minidx]
# Display the matching face
imshow2(test_img,
imtrain.images[minidx],
labels=('Test Image', 'Best Match: %d' % minidx))
P.title('L2 coefficient error: %.2e' % best_err)
def identify(test_img):
"""Try to find a match for test_img from the training set."""
# Normalise data, compute projections into eigenspace
XXX
# Find closest match
diff2 = ((proj_c - test_coeffs)**2).sum(axis=1) # this is correct.
minidx = # XXX Find the index of the minimum in diff2 (look for the argmin
# method)
best_err = # XXX And read the actual value off diff2 using this index:
# Display the matching face
imshow2(test_img,imtrain.images[minidx],
labels = ('Test Image','Best Match: %d' % minidx))
P.title('L2 coefficient error: %.2e' % best_err)
def verify(test_img,key):
"""Verify visually that a provided image corresponds to a specified image
in the training set.
Two images, the provided test image and the training image 'key', are
displayed side-by-side, along with the l2-norm error. Based on this
information, the user can visually verify that they are the same."""
# Compute the L2 error between the coefficients of the reference image:
ref_coeffs = proj_c[key]
# and the coefficients for your test image, which you must first normalize
# by removing data_mean from the flattened version of test_img:
norm_test = # XXX
test_coeffs = # XXX
l2_err = # XXX Look at N.linalg.norm for L2 vector norms
imshow2(imtrain[key],test_img,
labels = ('Reference Image','Test Image'))
P.title('L2 coefficient error: %.2e' % l2_err)
