def test_motion_blur(x, y, models):
# initialize histories
histories = init_histories({'kernel_size': mb_ksizes})
# run tests
for ksize in mb_ksizes:
# apply motion blur
print('[INFO] Applying motion blur with kernel size=({}, {})'\
.format(ksize, ksize))
noisy = []
count = 0
mb_kernel = numpy.zeros((ksize, ksize))
mb_kernel[ksize // 2, :] = 1
mb_kernel /= numpy.sum(mb_kernel)
for image in x:
noisy.append(
imdegrade(image,
'motion_blur',
mb_kernel=mb_kernel,
seed=RANDOM_SEED))
count += 1
print('\r[INFO] Progress... {:3.0f}%'\
.format(count*100/len(x)), end='')
# save noisy image samples if required
if SAVE_NOISY:
save_samples(noisy,
SAMP_NOISY,
randomize=True,
seed=RANDOM_SEED,
filename_prefix='motion_blur_ksize_({}x{})'.format(
ksize, ksize),
target_directory=OUTPUT_DIR_NOISY)
# test models
results = test_models(noisy, y, models)
# update histories
for hist_dict, res_dict in zip(histories, results):
for key in res_dict.keys():
hist_dict[key].append(res_dict[key])
# save and plot histories
save_and_plot_histories(file_id='motion_blur',
histories=histories,
title='Change in accuracy with motion blur',
xlabel='kernel size',
ylabel='accuracy (\u0025)')
return
def test_gaussian_color(x, y, models):
# initialize histories
histories = init_histories({'sigma': sigmavals})
# run tests
for sigma in sigmavals:
# apply gaussian color noise
print('[INFO] Applying Gaussian color noise with mu=0 and sigma={}'\
.format(sigma))
noisy = []
count = 0
for image in x:
noisy.append(
imdegrade(image,
'gaussian_color',
mu=0,
sigma=sigma,
seed=RANDOM_SEED))
count += 1
print('\r[INFO] Progress... {:3.0f}%'\
.format(count*100/len(x)), end='')
# save noisy image samples if required
if SAVE_NOISY:
save_samples(
noisy,
SAMP_NOISY,
randomize=True,
seed=RANDOM_SEED,
filename_prefix='gaussian_color_sigma_{}'.format(sigma),
target_directory=OUTPUT_DIR_NOISY)
# test models
results = test_models(noisy, y, models)
# update histories
for hist_dict, res_dict in zip(histories, results):
for key in res_dict.keys():
hist_dict[key].append(res_dict[key])
# save and plot histories
save_and_plot_histories(
file_id='gaussian_color',
histories=histories,
title='Change in accuracy with Gaussian color noise',
xlabel='standard deviation (\u03c3)',
ylabel='accuracy (\u0025)')
return
def test_salt_and_pepper(x, y, models):
# initialize histories
histories = init_histories({'density': densities})
# run tests
for density in densities:
# apply salt and pepper noise
print('[INFO] Applying salt and pepper noise with density={}'\
.format(density))
noisy = []
count = 0
for image in x:
noisy.append(
imdegrade(image,
'salt_and_pepper',
density=density,
seed=RANDOM_SEED))
count += 1
print('\r[INFO] Progress... {:3.0f}%'\
.format(count*100/len(x)), end='')
# save noisy image samples if required
if SAVE_NOISY:
save_samples(
noisy,
SAMP_NOISY,
randomize=True,
seed=RANDOM_SEED,
filename_prefix='salt_and_pepper_density_{}'.format(density),
target_directory=OUTPUT_DIR_NOISY)
# test models
results = test_models(noisy, y, models)
# update histories
for hist_dict, res_dict in zip(histories, results):
for key in res_dict.keys():
hist_dict[key].append(res_dict[key])
# save and plot histories
save_and_plot_histories(
file_id='salt_and_pepper',
histories=histories,
title='Change in accuracy with salt and pepper noise',
xlabel='noise density',
ylabel='accuracy (\u0025)')
return
def test_jpeg_quality(x, y, models):
# initialize histories
histories = init_histories({'image_quality': qualities})
# run tests
for quality in qualities:
# apply jpeg compression
print('[INFO] Applying JPEG compression with quality={}'\
.format(quality))
noisy = []
count = 0
for image in x:
noisy.append(
imdegrade(image,
'jpeg_compression',
quality=quality,
seed=RANDOM_SEED))
count += 1
print('\r[INFO] Progress... {:3.0f}%'\
.format(count*100/len(x)), end='')
# save noisy image samples if required
if SAVE_NOISY:
save_samples(noisy,
SAMP_NOISY,
randomize=True,
seed=RANDOM_SEED,
filename_prefix='jpeg_quality_{}'.format(quality),
target_directory=OUTPUT_DIR_NOISY)
# test models
results = test_models(noisy, y, models)
# update histories
for hist_dict, res_dict in zip(histories, results):
for key in res_dict.keys():
hist_dict[key].append(res_dict[key])
# save and plot histories
save_and_plot_histories(file_id='jpeg_quality',
histories=histories,
title='Change in accuracy with JPEG quality',
xlabel='image quality',
ylabel='accuracy (\u0025)',
invert_xaxis=True)
return