import matplotlib.pyplot as plt
from supreme.io import imread
from supreme.noise import dwt_denoise
import demo_data
X = demo_data.chelsea()
noise = np.random.normal(loc=0, scale=30, size=X.shape)
X_ = X + noise
X_ = np.clip(X_, 0, 255)
X_ /= X_.max()
Y = np.clip(dwt_denoise(X_, wavelet='db16', alpha=0.02, levels=4), 0, 255)
Y /= Y.max()
print "MSE:", np.mean((X - Y)**2)/np.prod(X.shape)
plt.subplot(2, 2, 1)
plt.imshow(X)
plt.title('Input Image')
plt.subplot(2, 2, 2)
plt.imshow(X_)
plt.title('Severe Noise Added')
plt.subplot(2, 2, 3)
plt.imshow(Y)
plt.title('Wavelet Filtered')
# --------------------------------------
alpha = 2
# --------------------------------------
if len(sys.argv) == 1:
print "Usage: noise_fingerprint.py file1 file2 file3 ..."
ic = ImageCollection(sys.argv[1:10])
n = []
denoise = []
for i, img in enumerate(ic):
print "Denoising image %d..." % i
clean = dwt_denoise(img, wavelet='db8', alpha=alpha, levels=4)
if i == 0:
denoise.append(clean)
n.append(img - clean)
avg = np.zeros(n[0].shape, dtype=float)
for noise in n:
avg += noise
avg /= len(n)
print "Noise minimum and maximum:", avg.min(), avg.max()
imdata = avg + avg.min()
imdata = imdata / imdata.max()