def bm3d_1st_step(sigma, img_noisy, nHard, kHard, NHard, pHard, lambdaHard3D,
tauMatch, useSD, tau_2D):
height, width = img_noisy.shape[0], img_noisy.shape[1]
row_ind = ind_initialize(height - kHard + 1, nHard, pHard)
column_ind = ind_initialize(width - kHard + 1, nHard, pHard)
kaiserWindow = get_kaiserWindow(kHard)
ri_rj_N__ni_nj, threshold_count = precompute_BM(img_noisy,
kHW=kHard,
NHW=NHard,
nHW=nHard,
tauMatch=tauMatch)
group_len = int(np.sum(threshold_count))
group_3D_table = np.zeros((group_len, kHard, kHard))
weight_table = np.zeros((height, width))
all_patches = image2patches(img_noisy, kHard,
kHard) # i_j_ipatch_jpatch__v
if tau_2D == 'DCT':
fre_all_patches = dct_2d_forward(all_patches)
else: # 'BIOR'
fre_all_patches = bior_2d_forward(all_patches)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
group_3D = build_3D_group(fre_all_patches,
ri_rj_N__ni_nj[i_r, j_r], nSx_r)
group_3D, weight = ht_filtering_hadamard(group_3D, sigma,
lambdaHard3D, not useSD)
group_3D = group_3D.transpose((2, 0, 1))
group_3D_table[acc_pointer:acc_pointer + nSx_r] = group_3D
acc_pointer += nSx_r
if useSD:
weight = sd_weighting(group_3D)
weight_table[i_r, j_r] = weight
if tau_2D == 'DCT':
group_3D_table = dct_2d_reverse(group_3D_table)
else: # 'BIOR'
group_3D_table = bior_2d_reverse(group_3D_table)
# group_3D_table = np.maximum(group_3D_table, 0)
# for i in range(1000):
# patch = group_3D_table[i]
# print(i, '----------------------------')
# print(patch)
# print(np.min(patch))
# print(np.max(patch))
# print(np.sum(patch))
# cv2.imshow('', patch.astype(np.uint8))
# cv2.waitKey()
numerator = np.zeros_like(img_noisy, dtype=np.float64)
denominator = np.zeros(
(img_noisy.shape[0] - 2 * nHard, img_noisy.shape[1] - 2 * nHard),
dtype=np.float64)
denominator = np.pad(denominator, nHard, 'constant', constant_values=1.)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
N_ni_nj = ri_rj_N__ni_nj[i_r, j_r]
group_3D = group_3D_table[acc_pointer:acc_pointer + nSx_r]
acc_pointer += nSx_r
weight = weight_table[i_r, j_r]
for n in range(nSx_r):
ni, nj = N_ni_nj[n]
patch = group_3D[n]
numerator[ni:ni + kHard,
nj:nj + kHard] += patch * kaiserWindow * weight
denominator[ni:ni + kHard,
nj:nj + kHard] += kaiserWindow * weight
img_basic = numerator / denominator
return img_basic
def bm3d_2nd_step(sigma, img_noisy, img_basic, nWien, kWien, NWien, pWien,
tauMatch, useSD, tau_2D, lam):
height, width = img_noisy.shape[0], img_noisy.shape[1]
row_ind = ind_initialize(height - kWien + 1, nWien, pWien)
column_ind = ind_initialize(width - kWien + 1, nWien, pWien)
kaiserWindow = get_kaiserWindow(kWien)
ri_rj_N__ni_nj, threshold_count = precompute_BM(img_basic,
kHW=kWien,
NHW=NWien,
nHW=nWien,
tauMatch=tauMatch)
group_len = int(np.sum(threshold_count))
group_3D_table = np.zeros((group_len, kWien, kWien))
weight_table = np.ones((height, width))
noisy_patches = image2patches(img_noisy, k=kWien,
p=pWien) # i_j_ipatch_jpatch__v
basic_patches = image2patches(img_basic, k=kWien,
p=pWien) # i_j_ipatch_jpatch__v
if tau_2D == 'DCT':
fre_noisy_patches = dct_2d_forward(noisy_patches)
fre_basic_patches = dct_2d_forward(basic_patches)
elif tau_2D == 'BIOR': # 'BIOR'
fre_noisy_patches = bior_2d_forward(noisy_patches)
fre_basic_patches = bior_2d_forward(basic_patches)
else:
fre_noisy_patches = np.zeros_like(noisy_patches)
for i, patch in enumerate(noisy_patches):
fre_noisy_patches[i] = ptv.tv1_2d(patch, lam)
fre_basic_patches = np.zeros_like(basic_patches)
for i, patch in enumerate(basic_patches):
fre_basic_patches[i] = ptv.tv1_2d(patch, lam)
fre_basic_patches = fre_basic_patches.reshape(
(height - kWien + 1, height - kWien + 1, kWien, kWien))
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
group_3D_est = build_3D_group(fre_basic_patches,
ri_rj_N__ni_nj[i_r, j_r], nSx_r)
group_3D = group_3D_est
group_3D = group_3D.transpose((2, 0, 1))
group_3D_table[acc_pointer:acc_pointer + nSx_r] = group_3D
acc_pointer += nSx_r
if useSD:
weight = sd_weighting(group_3D)
weight_table[i_r, j_r] = weight
if tau_2D == 'DCT':
group_3D_table = dct_2d_reverse(group_3D_table)
elif tau_2D == 'BIOR': # 'BIOR'
group_3D_table = bior_2d_reverse(group_3D_table)
else:
pass
# for i in range(1000):
# patch = group_3D_table[i]
# print(i, '----------------------------')
# print(patch)
# cv2.imshow('', patch.astype(np.uint8))
# cv2.waitKey()
group_3D_table *= kaiserWindow
numerator = np.zeros_like(img_noisy, dtype=np.float64)
denominator = np.zeros_like(img_noisy, dtype=np.float64)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
if i_r == 264 and j_r == 264:
print()
nSx_r = threshold_count[i_r, j_r]
N_ni_nj = ri_rj_N__ni_nj[i_r, j_r]
group_3D = group_3D_table[acc_pointer:acc_pointer + nSx_r]
acc_pointer += nSx_r
weight = weight_table[i_r, j_r]
for n in range(nSx_r):
ni, nj = N_ni_nj[n]
patch = group_3D[n]
numerator[ni:ni + kWien, nj:nj + kWien] += patch * weight
denominator[ni:ni + kWien,
nj:nj + kWien] += kaiserWindow * weight
img_denoised = numerator / denominator
return img_denoised
def bm3d_2nd_step_with_TV(sigma, img_noisy, img_basic, nWien, kWien, NWien,
pWien, tauMatch, useSD, tau_2D, lamb):
height, width = img_noisy.shape[0], img_noisy.shape[1]
row_ind = ind_initialize(height - kWien + 1, nWien, pWien)
column_ind = ind_initialize(width - kWien + 1, nWien, pWien)
kaiserWindow = get_kaiserWindow(kWien)
ri_rj_N__ni_nj, threshold_count = precompute_BM(img_basic,
kHW=kWien,
NHW=NWien,
nHW=nWien,
tauMatch=tauMatch)
group_len = int(np.sum(threshold_count))
group_3D_table = np.zeros((group_len, kWien, kWien))
weight_table = np.ones((height, width))
noisy_patches = image2patches(img_noisy, k=kWien,
p=pWien) # i_j_ipatch_jpatch__v
basic_patches = image2patches(img_basic, k=kWien,
p=pWien) # i_j_ipatch_jpatch__v
noisy_patches = noisy_patches.reshape(
(height - kWien + 1, height - kWien + 1, kWien, kWien))
basic_patches = basic_patches.reshape(
(height - kWien + 1, height - kWien + 1, kWien, kWien))
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
group_3D_est = build_3D_group(noisy_patches,
ri_rj_N__ni_nj[i_r, j_r], nSx_r)
group_3D = trendFilter3D(group_3D_est, lamb)
group_3D = group_3D.transpose((2, 0, 1))
group_3D_table[acc_pointer:acc_pointer + nSx_r] = group_3D
acc_pointer += nSx_r
# if useSD:
weight = sd_weighting(group_3D)
weight_table[i_r, j_r] = weight
group_3D_table *= kaiserWindow
numerator = np.zeros_like(img_noisy, dtype=np.float64)
denominator = np.zeros_like(img_noisy, dtype=np.float64)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
N_ni_nj = ri_rj_N__ni_nj[i_r, j_r]
group_3D = group_3D_table[acc_pointer:acc_pointer + nSx_r]
acc_pointer += nSx_r
weight = weight_table[i_r, j_r]
for n in range(nSx_r):
ni, nj = N_ni_nj[n]
patch = group_3D[n]
numerator[ni:ni + kWien, nj:nj + kWien] += patch * weight
denominator[ni:ni + kWien,
nj:nj + kWien] += kaiserWindow * weight
img_denoised = numerator / denominator
return img_denoised
def bm3d_2nd_step(sigma, img_noisy, img_basic, nWien, kWien, NWien, pWien,
tauMatch, useSD, tau_2D):
height, width = img_noisy.shape[0], img_noisy.shape[1]
row_ind = ind_initialize(height - kWien + 1, nWien, pWien)
column_ind = ind_initialize(width - kWien + 1, nWien, pWien)
kaiserWindow = get_kaiserWindow(kWien)
ri_rj_N__ni_nj, threshold_count = precompute_BM(img_basic,
kHW=kWien,
NHW=NWien,
nHW=nWien,
tauMatch=tauMatch)
group_len = int(np.sum(threshold_count))
group_3D_table = np.zeros((group_len, kWien, kWien))
weight_table = np.zeros((height, width))
noisy_patches = image2patches(img_noisy, kWien,
kWien) # i_j_ipatch_jpatch__v
basic_patches = image2patches(img_basic, kWien,
kWien) # i_j_ipatch_jpatch__v
if tau_2D == 'DCT':
fre_noisy_patches = dct_2d_forward(noisy_patches)
fre_basic_patches = dct_2d_forward(basic_patches)
else: # 'BIOR'
fre_noisy_patches = bior_2d_forward(noisy_patches)
fre_basic_patches = bior_2d_forward(basic_patches)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
group_3D_img = build_3D_group(fre_noisy_patches,
ri_rj_N__ni_nj[i_r, j_r], nSx_r)
group_3D_est = build_3D_group(fre_basic_patches,
ri_rj_N__ni_nj[i_r, j_r], nSx_r)
group_3D, weight = wiener_filtering_hadamard(
group_3D_img, group_3D_est, sigma, not useSD)
group_3D = group_3D.transpose((2, 0, 1))
group_3D_table[acc_pointer:acc_pointer + nSx_r] = group_3D
acc_pointer += nSx_r
if useSD:
weight = sd_weighting(group_3D)
weight_table[i_r, j_r] = weight
if tau_2D == 'DCT':
group_3D_table = dct_2d_reverse(group_3D_table)
else: # 'BIOR'
group_3D_table = bior_2d_reverse(group_3D_table)
# for i in range(1000):
# patch = group_3D_table[i]
# print(i, '----------------------------')
# print(patch)
# cv2.imshow('', patch.astype(np.uint8))
# cv2.waitKey()
# aggregation part
numerator = np.zeros_like(img_noisy, dtype=np.float64)
denominator = np.zeros(
(img_noisy.shape[0] - 2 * nWien, img_noisy.shape[1] - 2 * nWien),
dtype=np.float64)
denominator = np.pad(denominator, nWien, 'constant', constant_values=1.)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
N_ni_nj = ri_rj_N__ni_nj[i_r, j_r]
group_3D = group_3D_table[acc_pointer:acc_pointer + nSx_r]
acc_pointer += nSx_r
weight = weight_table[i_r, j_r]
for n in range(nSx_r):
ni, nj = N_ni_nj[n]
patch = group_3D[n]
numerator[ni:ni + kWien,
nj:nj + kWien] += patch * kaiserWindow * weight
denominator[ni:ni + kWien,
nj:nj + kWien] += kaiserWindow * weight
img_denoised = numerator / denominator
return img_denoised
def bm3d_1st_step_with_TV(sigma, img_noisy, nHard, kHard, NHard, pHard, lambdaHard3D, tauMatch, useSD, tau_2D, lamb):
height, width = img_noisy.shape[0], img_noisy.shape[1]
row_ind = ind_initialize(height - kHard + 1, nHard, pHard)
column_ind = ind_initialize(width - kHard + 1, nHard, pHard)
kaiserWindow = get_kaiserWindow(kHard)
ri_rj_N__ni_nj, threshold_count = precompute_BM(img_noisy, kHW=kHard, NHW=NHard, nHW=nHard, tauMatch=tauMatch)
for i in range(height):
for j in range(width):
assert ri_rj_N__ni_nj[i][j][0][0] == i
assert ri_rj_N__ni_nj[i][j][0][1] == j
group_len = int(np.sum(threshold_count))
group_3D_table = np.zeros((group_len, kHard, kHard))
weight_table = np.zeros((height, width))
all_patches = image2patches(img_noisy, k=kHard, p=pHard) # i_j_ipatch_jpatch__v
if tau_2D == 'DCT':
fre_all_patches = dct_2d_forward(all_patches)
elif tau_2D == 'BIOR': # 'BIOR'
fre_all_patches = bior_2d_forward(all_patches)
else:
fre_all_patches = all_patches
fre_all_patches = fre_all_patches.reshape((height - kHard + 1, height - kHard + 1, kHard, kHard))
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
group_3D = build_3D_group(fre_all_patches, ri_rj_N__ni_nj[i_r, j_r], nSx_r)
# group_3D_temp, weight = ht_filtering_hadamard(group_3D, sigma, lambdaHard3D, not useSD)
group_3D = trendFilter3D(group_3D, lamb)
group_3D = group_3D.transpose((2, 0, 1))
group_3D_table[acc_pointer:acc_pointer + nSx_r] = group_3D
acc_pointer += nSx_r
if useSD:
weight = sd_weighting(group_3D)
weight_table[i_r, j_r] = weight
if tau_2D == 'DCT':
group_3D_table = dct_2d_reverse(group_3D_table)
elif tau_2D == 'BIOR': # 'BIOR'
group_3D_table = bior_2d_reverse(group_3D_table)
else:
group_3D_table *= kaiserWindow
numerator = np.zeros_like(img_noisy, dtype=np.float64)
denominator = np.zeros((img_noisy.shape[0] - 2 * nHard, img_noisy.shape[1] - 2 * nHard), dtype=np.float64)
denominator = np.pad(denominator, nHard, 'constant', constant_values=1.)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
N_ni_nj = ri_rj_N__ni_nj[i_r, j_r]
group_3D = group_3D_table[acc_pointer:acc_pointer + nSx_r]
acc_pointer += nSx_r
weight = weight_table[i_r, j_r]
for n in range(nSx_r):
ni, nj = N_ni_nj[n]
patch = group_3D[n]
if tau_2D == 'DCT' or tau_2D == 'BIOR':
numerator[ni:ni + kHard, nj:nj + kHard] += patch * kaiserWindow * weight
denominator[ni:ni + kHard, nj:nj + kHard] += kaiserWindow * weight
elif tau_2D == 'TV':
numerator[ni:ni + kHard, nj:nj + kHard] += patch * weight
denominator[ni:ni + kHard, nj:nj + kHard] += kaiserWindow * weight
img_basic = numerator / denominator
return img_basic
def bm3d_1st_step(sigma, img_noisy, nHard, kHard, NHard, pHard, useSD, tau_2D):
height, width = img_noisy.shape[0], img_noisy.shape[1]
lambdaHard3D = 2.7 # ! Threshold for Hard Thresholding
tauMatch = 3 * (2500 if sigma < 35 else 5000
) # ! threshold determinates similarity between patches
row_ind = ind_initialize(height - kHard + 1, nHard, pHard)
column_ind = ind_initialize(width - kHard + 1, nHard, pHard)
kaiserWindow = get_kaiserWindow(kHard)
ri_rj_N__ni_nj, threshold_count = precompute_BM(img_noisy,
kHW=kHard,
NHW=NHard,
nHW=nHard,
tauMatch=tauMatch)
group_len = int(np.sum(threshold_count))
group_3D_table = np.zeros((group_len, kHard, kHard))
weight_table = np.zeros((height, width))
all_patches = image2patches(img_noisy, k=kHard,
p=pHard) # i_j_ipatch_jpatch__v
if tau_2D == 'DCT':
fre_all_patches = dct_2d_forward(all_patches)
else: # 'BIOR'
fre_all_patches = bior_2d_forward(all_patches)
fre_all_patches = fre_all_patches.reshape(
(height - kHard + 1, height - kHard + 1, kHard, kHard))
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
group_3D = build_3D_group(fre_all_patches,
ri_rj_N__ni_nj[i_r, j_r], nSx_r)
group_3D = group_3D.reshape(kHard * kHard, nSx_r)
# group_3D, weight = ht_filtering_hadamard(group_3D, sigma, lambdaHard3D, not useSD)
group_3D = group_3D.reshape(kHard, kHard, nSx_r)
group_3D = group_3D.transpose((2, 0, 1))
group_3D_table[acc_pointer:acc_pointer + nSx_r] = group_3D
acc_pointer += nSx_r
if useSD:
weight = sd_weighting(group_3D)
# weight_table[i_r, j_r] = weight
weight_table[i_r, j_r] = 1
if tau_2D == 'DCT':
group_3D_table = dct_2d_reverse(group_3D_table)
else: # 'BIOR'
group_3D_table = bior_2d_reverse(group_3D_table)
# for i in range(1000):
# patch = group_3D_table[i]
# print(i, '----------------------------')
# print(patch)
# cv2.imshow('', patch.astype(np.uint8))
# cv2.waitKey()
group_3D_table *= kaiserWindow
numerator = np.zeros_like(img_noisy, dtype=np.float)
denominator = np.zeros_like(img_noisy, dtype=np.float)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
N_ni_nj = ri_rj_N__ni_nj[i_r, j_r]
group_3D = group_3D_table[acc_pointer:acc_pointer + nSx_r]
acc_pointer += nSx_r
weight = weight_table[i_r, j_r]
for n in range(nSx_r):
ni, nj = N_ni_nj[n]
patch = group_3D[n]
numerator[ni:ni + kHard, nj:nj + kHard] += patch * weight
denominator[ni:ni + kHard,
nj:nj + kHard] += kaiserWindow * weight
img_basic = numerator / denominator
img_basic = img_basic.astype(np.uint8)
return img_basic
def bm3d_1st_step_block_mean(sigma, img_noisy, nHard, kHard, NHard, pHard, lambdaHard3D, tauMatch, useSD, tau_2D,
block_mean=False):
height, width = img_noisy.shape[0], img_noisy.shape[1]
row_ind = ind_initialize(height - kHard + 1, nHard, pHard)
column_ind = ind_initialize(width - kHard + 1, nHard, pHard)
kaiserWindow = get_kaiserWindow(kHard)
ri_rj_N__ni_nj, threshold_count = precompute_BM(img_noisy, kHW=kHard, NHW=NHard, nHW=nHard, tauMatch=tauMatch)
group_len = int(np.sum(threshold_count))
group_3D_table = np.zeros((group_len, kHard, kHard))
weight_table = np.zeros((height, width))
all_patches = image2patches(img_noisy, k=kHard, p=pHard) # i_j_ipatch_jpatch__v
if tau_2D == 'DCT':
fre_all_patches = dct_2d_forward(all_patches)
else: # 'BIOR'
fre_all_patches = bior_2d_forward(all_patches)
fre_all_patches = fre_all_patches.reshape((height - kHard + 1, height - kHard + 1, kHard, kHard))
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
group_3D = build_3D_group(fre_all_patches, ri_rj_N__ni_nj[i_r, j_r], nSx_r)
if block_mean:
group_3D, weight = get_mean(group_3D)
else:
group_3D, weight = ht_filtering_hadamard(group_3D, sigma, lambdaHard3D, not useSD)
group_3D = group_3D.transpose((2, 0, 1))
group_3D_table[acc_pointer:acc_pointer + nSx_r] = group_3D
acc_pointer += nSx_r
if useSD:
weight = sd_weighting(group_3D)
weight_table[i_r, j_r] = weight
if tau_2D == 'DCT':
group_3D_table = dct_2d_reverse(group_3D_table)
else: # 'BIOR'
group_3D_table = bior_2d_reverse(group_3D_table)
numerator = np.zeros_like(img_noisy, dtype=np.float64)
denominator = np.zeros_like(img_noisy, dtype=np.float64)
acc_pointer = 0
for i_r in row_ind:
for j_r in column_ind:
nSx_r = threshold_count[i_r, j_r]
N_ni_nj = ri_rj_N__ni_nj[i_r, j_r]
group_3D = group_3D_table[acc_pointer:acc_pointer + nSx_r]
acc_pointer += nSx_r
weight = weight_table[i_r, j_r]
for n in range(nSx_r):
ni, nj = N_ni_nj[n]
patch = group_3D[n]
numerator[ni:ni + kHard, nj:nj + kHard] += patch * kaiserWindow * weight
denominator[ni:ni + kHard, nj:nj + kHard] += kaiserWindow * weight
img_basic = numerator / denominator
return img_basic
