def maximum_noise_plot():
''' Generates a plot showing the effectiveness of the Loopy BP
algorithm for varying mu values, used in the write-up.
'''
disk_fp = image_processing.generate_disk(100, 40)
original_image = Image(disk_fp)
FLIP_PROBS = [.005, .01, .05, .1, .2, .3, .4, .5]
noises = []
for prob in FLIP_PROBS:
noisy_fp = image_processing.noisify(disk_fp, prob)
# plt.imshow(np.genfromtxt(noisy_fp), cmap='Greys')
# plt.show()
noisy_image = Image(noisy_fp)
denoised_fp = loopy_bp(noisy_image, THETA)
denoised_image = Image(denoised_fp)
fraction_noise = original_image.percent_different(denoised_image)
print fraction_noise
noises.append(fraction_noise)
plt.plot(FLIP_PROBS, noises)
plt.plot(FLIP_PROBS, FLIP_PROBS)
plt.xlabel("Mu value (flip probability)")
plt.ylabel("Fraction noise")
plt.show()
def find_optimal_theta():
''' Produces the plot finding the optimal Theta, used in the write-up.
I found that the optimal thetas were the highest, and I decided to use
0.9 for the bulk of my analysis.
'''
THETAS = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.999]
FLIP_PROBS = [0.005, 0.05, 0.25, 0.5]
noises = []
disk_fp = image_processing.generate_disk(100, 40)
original_image = Image(disk_fp)
noisy_fps = []
for prob in FLIP_PROBS:
noisy_fps.append(image_processing.noisify(disk_fp, prob))
for noisy_image_fp in noisy_fps:
noise = []
for theta in THETAS:
print noisy_image_fp, theta
noisy_image = Image(noisy_image_fp)
loopy_bp(noisy_image, theta)
denoised_image = Image(noisy_image.file_path + "_denoised")
percent_noise = original_image.percent_different(denoised_image)
noise.append(percent_noise)
os.remove(noisy_image.file_path + "_denoised")
noises.append(noise)
for noise in noises:
plt.plot(THETAS, noise)
plt.xlabel("Theta")
plt.ylabel("Fraction noise")
plt.show()
def print_images(): ''' Prints all the images shown in the write-up.''' # Circle with radius 40 disk_fp = image_processing.generate_disk(100, 40) show_image(disk_fp) # Circle with mu = 0.01 noisy_fp = image_processing.noisify(disk_fp, 0.01) show_image(noisy_fp) # Circle cleaned up, mu = 0.01 noisy_image = Image(noisy_fp) denoised_fp = loopy_bp(noisy_image, THETA) show_image(denoised_fp) # Circle with mu = 0.1 noisy_fp = image_processing.noisify(disk_fp, 0.1) show_image(noisy_fp) # Circle cleaned up, mu = 0.1 noisy_image = Image(noisy_fp) denoised_fp = loopy_bp(noisy_image, THETA) show_image(denoised_fp) # Square, 60 x 60 square_fp = image_processing.generate_square(100, 60, 50) show_image(square_fp) # Square with mu = 0.01 noisy_fp = image_processing.noisify(square_fp, 0.01) show_image(noisy_fp) # Square cleaned up, mu = 0.01 noisy_image = Image(noisy_fp) denoised_fp = loopy_bp(noisy_image, THETA) show_image(denoised_fp) # Equals Sign equals_fp = image_processing.generate_equalsign() show_image(equals_fp) # Equals with mu = 0.01 noisy_fp = image_processing.noisify(equals_fp, 0.01) show_image(noisy_fp) # Square cleaned up, mu = 0.01 noisy_image = Image(noisy_fp) denoised_fp = loopy_bp(noisy_image, THETA) show_image(denoised_fp)