def color_retinex(image, mask, threshold_gray, threshold_color, L1=False):
image = np.clip(image, 3., np.infty)
log_image = np.log(image)
i_y_orig, i_x_orig = poisson.get_gradients(log_image)
i_y_gray, i_y_color = project_gray(i_y_orig), project_chromaticity(i_y_orig)
i_x_gray, i_x_color = project_gray(i_x_orig), project_chromaticity(i_x_orig)
image_grayscale = np.mean(image, axis=2)
image_grayscale = np.clip(image_grayscale, 3., np.infty)
log_image_grayscale = np.log(image_grayscale)
i_y, i_x = poisson.get_gradients(log_image_grayscale)
norm = np.sqrt(np.sum(i_y_color**2, axis=2))
i_y_match = (norm > threshold_color) + (np.abs(i_y_gray[:,:,0]) > threshold_gray)
norm = np.sqrt(np.sum(i_x_color**2, axis=2))
i_x_match = (norm > threshold_color) + (np.abs(i_x_gray[:,:,0]) > threshold_gray)
r_y = np.where(i_y_match, i_y, 0.)
r_x = np.where(i_x_match, i_x, 0.)
if L1:
log_refl = poisson.solve_L1(r_y, r_x, mask)
else:
log_refl = poisson.solve(r_y, r_x, mask)
refl = np.exp(log_refl)
return image_grayscale / refl, refl
def color_retinex(image, mask, threshold_gray, threshold_color, L1=False):
# Clip the values of image to lie in the range of [3,infty)
image = np.clip(image, 3., np.infty)
log_image = np.log(image)
# Computes x and y gradients using [1,-1] filters, see poisson.py.
i_y_orig, i_x_orig = poisson.get_gradients(log_image)
i_y_gray, i_y_color = project_gray(i_y_orig), project_chromaticity(i_y_orig)
i_x_gray, i_x_color = project_gray(i_x_orig), project_chromaticity(i_x_orig)
image_grayscale = np.mean(image, axis=2)
image_grayscale = np.clip(image_grayscale, 3., np.infty)
log_image_grayscale = np.log(image_grayscale)
i_y, i_x = poisson.get_gradients(log_image_grayscale)
norm = np.sqrt(np.sum(i_y_color**2, axis=2))
i_y_match = (norm > threshold_color) + (np.abs(i_y_gray[:,:,0]) > threshold_gray)
norm = np.sqrt(np.sum(i_x_color**2, axis=2))
i_x_match = (norm > threshold_color) + (np.abs(i_x_gray[:,:,0]) > threshold_gray)
r_y = np.where(i_y_match, i_y, 0.)
r_x = np.where(i_x_match, i_x, 0.)
if L1:
log_refl = poisson.solve_L1(r_y, r_x, mask)
else:
log_refl = poisson.solve(r_y, r_x, mask)
refl = np.exp(log_refl)
return image_grayscale / refl, refl
def color_retinex(image, mask, threshold_gray, threshold_color, L1=False):
image = np.clip(image, 3., np.infty)
log_image = np.log(image)
i_y_orig, i_x_orig = poisson.get_gradients(log_image)
i_y_gray, i_y_color = project_gray(i_y_orig), project_chromaticity(
i_y_orig)
i_x_gray, i_x_color = project_gray(i_x_orig), project_chromaticity(
i_x_orig)
image_grayscale = np.mean(image, axis=2)
image_grayscale = np.clip(image_grayscale, 3., np.infty)
log_image_grayscale = np.log(image_grayscale)
i_y, i_x = poisson.get_gradients(log_image_grayscale)
norm = np.sqrt(np.sum(i_y_color**2, axis=2))
i_y_match = (norm > threshold_color) + (np.abs(i_y_gray[:, :, 0]) >
threshold_gray)
norm = np.sqrt(np.sum(i_x_color**2, axis=2))
i_x_match = (norm > threshold_color) + (np.abs(i_x_gray[:, :, 0]) >
threshold_gray)
r_y = np.where(i_y_match, i_y, 0.)
r_x = np.where(i_x_match, i_x, 0.)
if L1:
log_refl = poisson.solve_L1(r_y, r_x, mask)
else:
log_refl = poisson.solve(r_y, r_x, mask)
refl = np.exp(log_refl)
return image_grayscale / refl, refl
def weiss(image, multi_images, mask, L1=False):
multi_images = np.clip(multi_images, 3., np.infty)
log_multi_images = np.log(multi_images)
i_y_all, i_x_all = poisson.get_gradients(log_multi_images)
r_y = np.median(i_y_all, axis=2)
r_x = np.median(i_x_all, axis=2)
if L1:
log_refl = poisson.solve_L1(r_y, r_x, mask)
else:
log_refl = poisson.solve(r_y, r_x, mask)
refl = np.where(mask, np.exp(log_refl), 0.)
shading = np.where(mask, image / refl, 0.)
return shading, refl
def retinex(image, mask, threshold, L1=False):
image = np.clip(image, 3., np.infty)
log_image = np.where(mask, np.log(image), 0.)
i_y, i_x = poisson.get_gradients(log_image)
r_y = np.where(np.abs(i_y) > threshold, i_y, 0.)
r_x = np.where(np.abs(i_x) > threshold, i_x, 0.)
if L1:
log_refl = poisson.solve_L1(r_y, r_x, mask)
else:
log_refl = poisson.solve(r_y, r_x, mask)
refl = mask * np.exp(log_refl)
return np.where(mask, image / refl, 0.), refl
def weiss(image, multi_images, mask, L1=False):
multi_images = np.clip(multi_images, 3., np.infty)
log_multi_images = np.log(multi_images)
i_y_all, i_x_all = poisson.get_gradients(log_multi_images)
r_y = np.median(i_y_all, axis=2)
r_x = np.median(i_x_all, axis=2)
if L1:
log_refl = poisson.solve_L1(r_y, r_x, mask)
else:
log_refl = poisson.solve(r_y, r_x, mask)
refl = np.where(mask, np.exp(log_refl), 0.)
shading = np.where(mask, image / refl, 0.)
return shading, refl
def retinex(image, mask, threshold, L1=False):
image = np.clip(image, 3., np.infty)
log_image = np.where(mask, np.log(image), 0.)
i_y, i_x = poisson.get_gradients(log_image)
r_y = np.where(np.abs(i_y) > threshold, i_y, 0.)
r_x = np.where(np.abs(i_x) > threshold, i_x, 0.)
if L1:
log_refl = poisson.solve_L1(r_y, r_x, mask)
else:
log_refl = poisson.solve(r_y, r_x, mask)
refl = mask * np.exp(log_refl)
return np.where(mask, image / refl, 0.), refl
n = grid * mul psi = np.zeros((n, n), dtype=np.float32) ome = np.zeros((n, n), dtype=np.float32) dt = float(0.01 / mul) nu = float(0.061) ps = psi om = ome ps_list = [] #om_list = [] for t in range(int(1 / dt)): # omega を時間発展する omout = evolve.evolve(ps, om, dt, nu) # psi を poisson 方程式から求める psout = poisson.solve(ps, omout) ps_list.append(psout) # om_list.append(omout) ps = psout om = omout kx = np.array([ [-1, 1] ]) ky = np.array([ [-1], [1] ]) xx, yy = np.meshgrid(np.linspace(0.0, 1.0, n), np.linspace(0.0, 1.0, n)) xs = xx[0::mul, 0::mul] ys = yy[0::mul, 0::mul] fig = plt.figure(figsize=(10,10)) ims = [] for t in range(4, len(ps_list), 5): u = signal.correlate(ps_list[t], ky, 'same') * n v = signal.correlate(ps_list[t], -kx, 'same') * n