def blend_color(self,
refered_pixel_coordi_lst,
blending_area_mask,
sigma1=0.5,
sigma2=0.5):
overlap_slic = MaskedSLIC(self.tar_img,
blending_area_mask,
region_size=20)
sampling_pixel_coordi_lst = np.array([
(rows[len(rows) // 2], cols[len(rows) // 2])
for idx, (rows, cols) in enumerate(overlap_slic.labels_position)
if idx != 0
])
color_compensation_lst = self.__calc_color_compensation(
sampling_pixel_coordi_lst, refered_pixel_coordi_lst, sigma1,
sigma2)
ref_img = np.copy(self.ref_img)
blended_img = np.zeros(self.ref_img.shape, dtype=np.float64)
tar_img_norm = self.tar_img / 255
for idx, color_compensation in enumerate(color_compensation_lst):
super_pixel_coordis = overlap_slic.labels_position[idx + 1]
blended_img[super_pixel_coordis[0], super_pixel_coordis[1]] = (
(tar_img_norm[super_pixel_coordis[0], super_pixel_coordis[1]] +
color_compensation) * 255)
blended_img[blended_img < 0] = 0
blended_img[blended_img > 255] = 255
ref_img[blended_img > 0] = blended_img[blended_img > 0].astype(
np.uint8)
return ref_img
def main():
'''
Initialization
'''
start = time.time()
tar_4_corners_xy = None
ref_4_corners_xy = None
homoMat = np.load('image/save_H.npy')
shiftMat = np.load('image/save_Shift.npy')
ref_4_corners_xy = np.load('image/save_ref_4_corners_xy.npy')
tar_4_corners_xy = np.load('image/save_tar_4_corners_xy.npy')
warp_tar_img = cv2.imread('image/warped_target.png')
warp_ref_img = cv2.imread('image/warped_reference.png')
seam_mask = cv2.imread('image/seam_mask.png', cv2.IMREAD_GRAYSCALE)
mask = Mask(warp_tar_img, warp_ref_img)
ref_region_mask = cv2.imread('image/result_from_reference.png',
cv2.IMREAD_GRAYSCALE)
tar_region_mask = cv2.bitwise_and(cv2.bitwise_not(ref_region_mask),
mask.tar)
mask.tar_result = tar_region_mask
mask.ref_result = ref_region_mask
'''
Blend color
'''
CB = CBlender(warp_tar_img, warp_ref_img, seam_mask, tar_4_corners_xy,
ref_4_corners_xy, homoMat, shiftMat, mask)
devided_tar_nonoverlap_mask, devided_tar_edge_mask = CB.split_tar_nonoverlap_area_and_edge(
ref_4_corners_xy, tar_4_corners_xy, shiftMat, homoMat)
devided_tar_edge_mask_1 = np.copy(devided_tar_edge_mask)
devided_tar_edge_mask_1[devided_tar_edge_mask_1 == 127] = 0
devided_tar_edge_mask_2 = np.copy(devided_tar_edge_mask)
devided_tar_edge_mask_2[devided_tar_edge_mask_1 == 255] = 0
# refered_pixel_coordi_lst = np.vstack( (\
# np.array( np.nonzero(seam_mask) ).T, \
# np.array( np.where(devided_tar_edge_mask==255) ).T, \
# np.array( np.where(devided_tar_edge_mask==127) ).T \
# ))
slic = MaskedSLIC(warp_tar_img,
np.bitwise_and(mask.tar_result, mask.overlap),
region_size=20,
compactness=5)
seam_superpixel_idx_lst = np.array([
idx for idx, (rows, cols) in enumerate(slic.labels_position)
if np.sum(seam_mask[rows, cols]) > 0
])
tar_edge_1_superpixel_idx_lst = np.array([
idx for idx, (rows, cols) in enumerate(slic.labels_position)
if np.sum(devided_tar_edge_mask_1[rows, cols]) > 0
])
tar_edge_2_superpixel_idx_lst = np.array([
idx for idx, (rows, cols) in enumerate(slic.labels_position)
if np.sum(devided_tar_edge_mask_2[rows, cols]) > 0
])
refered_idx_lst = np.hstack(
(seam_superpixel_idx_lst, tar_edge_1_superpixel_idx_lst,
tar_edge_2_superpixel_idx_lst))
refered_pixel_coordi_lst = np.array([
(rows[len(rows) // 2], cols[len(rows) // 2])
for idx, (rows, cols) in enumerate(slic.labels_position)
if (idx != 0) and (np.isin(idx, refered_idx_lst))
])
fuck_this_shit = CB.blend_color(refered_pixel_coordi_lst,
np.bitwise_and(CB.mask.overlap,
CB.mask.tar_result),
sigma1=0.3,
sigma2=0.2)
cv2.imwrite('overlap_blending_only.png', fuck_this_shit)
CB.ref_img = fuck_this_shit
refered_pixel_coordi_lst = np.array([
(rows[len(rows) // 2], cols[len(rows) // 2])
for idx, (rows, cols) in enumerate(slic.labels_position)
if (idx != 0) and (np.isin(idx, tar_edge_1_superpixel_idx_lst))
])
mask = np.zeros(devided_tar_nonoverlap_mask.shape)
mask[devided_tar_nonoverlap_mask == 255] = 255
fuck_this_shit = CB.blend_color(refered_pixel_coordi_lst,
mask,
sigma1=0.1,
sigma2=0.05)
cv2.imwrite('nonoverlap_blending_only1.png', fuck_this_shit)
CB.ref_img = fuck_this_shit
refered_pixel_coordi_lst = np.array([
(rows[len(rows) // 2], cols[len(rows) // 2])
for idx, (rows, cols) in enumerate(slic.labels_position)
if (idx != 0) and (np.isin(idx, tar_edge_2_superpixel_idx_lst))
])
mask = np.zeros(devided_tar_nonoverlap_mask.shape)
mask[devided_tar_nonoverlap_mask == 127] = 255
fuck_this_shit = CB.blend_color(refered_pixel_coordi_lst,
mask,
sigma1=0.1,
sigma2=0.05)
cv2.imwrite('nonoverlap_blending_only2.png', fuck_this_shit)
for idx in tar_edge_2_superpixel_idx_lst:
fuck_this_shit[slic.labels_position[idx][0],
slic.labels_position[idx][1]] = np.random.randint(
0, 255, 3)
cv2.imwrite('araara.png', fuck_this_shit)
warp_ref_img[slic.contour_mask > 0] = (0, 255, 0)
warp_tar_img[slic.contour_mask > 0] = (0, 255, 0)
cv2.imwrite('target_image_slic.png', warp_ref_img)
cv2.imwrite('reference_image_slic.png', warp_tar_img)
print(f'time: {time.time() - start}')
start = time.time()
warp_tar_img = cv2.cvtColor(cv2.imread('image/warped_target.png'), cv2.COLOR_BGR2YUV)
warp_ref_img = cv2.cvtColor(cv2.imread('image/warped_reference.png'), cv2.COLOR_BGR2YUV)
mask = Mask(warp_tar_img, warp_ref_img)
seam_mask = cv2.imread('image/seam_mask.png',cv2.IMREAD_GRAYSCALE)
ref_region_mask = cv2.imread('image/result_from_reference.png',cv2.IMREAD_GRAYSCALE)
tar_region_mask = cv2.bitwise_and( cv2.bitwise_not(ref_region_mask) , mask.tar )
mask.tar_result = tar_region_mask
mask.ref_result = ref_region_mask
slic = MaskedSLIC(cv2.cvtColor(warp_tar_img, cv2.COLOR_YUV2BGR), np.bitwise_and(mask.tar_result, mask.overlap) ,region_size=20, compactness=5)
seam_superpixel_idx_lst = np.array([idx for idx, (rows, cols) in enumerate(slic.labels_position) if np.sum(seam_mask[rows, cols]) > 0])
# print(seam_superpixel_idx_lst)
median_tar_img = np.copy(warp_tar_img)
for i in range(len(slic.labels_position)):
if i != 0:
rows, cols = slic.labels_position[i]
median_tar_img[rows, cols] = np.mean(median_tar_img[rows, cols], axis=0)
median_ref_img = np.copy(warp_ref_img)
for i in range(len(slic.labels_position)):
if i != 0:
rows, cols = slic.labels_position[i]
median_ref_img[rows, cols] = np.mean(median_ref_img[rows, cols], axis=0)
start = time.time()
warp_tar_img = cv2.imread('image/warped_target.png')
warp_ref_img = cv2.imread('image/warped_reference.png')
mask = Mask(warp_tar_img, warp_ref_img)
seam_mask = cv2.imread('image/seam_mask.png', cv2.IMREAD_GRAYSCALE)
ref_region_mask = cv2.imread('image/result_from_reference.png',
cv2.IMREAD_GRAYSCALE)
tar_region_mask = cv2.bitwise_and(cv2.bitwise_not(ref_region_mask), mask.tar)
mask.tar_result = tar_region_mask
mask.ref_result = ref_region_mask
slic = MaskedSLIC(warp_tar_img,
np.bitwise_and(mask.tar_result, mask.overlap),
region_size=20,
compactness=3)
CB = CBlender(warp_tar_img, warp_ref_img, seam_mask, mask)
refered_pixel_coordi_lst = np.array([
(rows[len(rows) // 2], cols[len(rows) // 2])
for idx, (rows, cols) in enumerate(slic.labels_position) if idx != 0
])
sigma1 = 0.1
sigma2 = 0.25
# print(refered_pixel_cooordi_lst)
a = CB.blend_color(refered_pixel_coordi_lst, mask.tar_result, sigma1, sigma2)
# a[refered_pixel_cooordi_lst[0], refered_pixel_cooordi_lst[1],:] = (255,255,255)
# a[slic.labels_position[1][0], slic.labels_position[1][1]] = (0,0,255)
# a[slic.contour_mask>0] = (0,255,0)
d = labels.ravel()
f = lambda x: np.unravel_index(x.index, labels.shape)
return pd.Series(d).groupby(d).apply(f)
start_time = time.time()
tar_img = cv2.imread('./data/processed_image/warped_target.png')
ref_img = cv2.imread('./data/processed_image/warped_reference.png')
mask = Mask(tar_img, ref_img)
# segmentation = slic(tar_img, n_segments=9)#, mask=mask.overlap)
# g = graph.rag_mean_color(tar_img, segmentation)
# lc = graph.show_rag(segmentation, g ,tar_img)
# cbar = plt.colorbar(lc)
# numOfPixel = len(np.unique(segmentation))
maskedSLIC = MaskedSLIC(tar_img, mask.overlap, region_size=200)
tar_img[maskedSLIC.contour_mask > 0] = (0, 255, 0)
cv2.imwrite('slic_contour.png', tar_img)
g = graph.rag_mean_color(tar_img, maskedSLIC.labels)
fig, ax = plt.subplots(nrows=2, sharex=True, sharey=True, figsize=(6, 8))
ax[0].set_title('RAG drawn with default settings')
lc = graph.show_rag(labels, g, img, ax=ax[0])
# specify the fraction of the plot area that will be used to draw the colorbar
fig.colorbar(lc, fraction=0.03, ax=ax[0])
ax[1].set_title('RAG drawn with grayscale image and viridis colormap')
lc = graph.show_rag(labels,
g,
img,
from cv2 import cv2
import numpy as np
from mask import Mask
import time
from SLIC import MaskedSLIC
import pickle
start_time = time.time()
tar_img = cv2.imread('./data/processed_image/warped_target.png')
ref_img = cv2.imread('./data/processed_image/warped_reference.png')
mask = Mask(tar_img, ref_img)
maskedSLIC = MaskedSLIC(ref_img, mask.overlap, region_size=20)
# for i in maskedSLIC.adjacent_pairs[0]:
# print(i)
# for (rows, cols) in maskedSLIC.labels_position:
# tar_img[rows, cols] = np.random.randint(0,256,size=(1,3))
# cv2.imwrite('./slic_pandas.png',tar_img)
print(time.time() - start_time)
print(f'number of pixel is :{maskedSLIC.numOfPixel}')
with open('./data/SLIC/maskedSLIC.pkl', 'wb') as output:
pickle.dump(maskedSLIC, output, pickle.HIGHEST_PROTOCOL)
return ref_img
start = time.time()
warp_tar_img = cv2.imread('image/warped_target.png')
warp_ref_img = cv2.imread('image/warped_reference.png')
seam_mask = cv2.imread('image/seam_mask.png', cv2.IMREAD_GRAYSCALE)
mask = Mask(warp_tar_img, warp_ref_img)
ref_region_mask = cv2.imread('image/result_from_reference.png',
cv2.IMREAD_GRAYSCALE)
tar_region_mask = cv2.bitwise_and(cv2.bitwise_not(ref_region_mask), mask.tar)
maskSLIC = MaskedSLIC(warp_tar_img, tar_region_mask, region_size=20)
CB = CBlender(warp_tar_img, warp_ref_img, maskSLIC, seam_mask, tar_region_mask)
blended_img = CB.blend_color()
print(CB.maskSLIC.numOfPixel)
print(np.sum(CB.tar_mask) / 255)
print(f'time: {time.time()-start}')
cv2.imwrite('blended_img.png', blended_img)
blended_img[maskSLIC.contour_mask > 0] = (0, 255, 0)
cv2.imwrite('pixel_blended_img.png', blended_img)
blended_img[seam_mask > 0] = (0, 0, 255)
cv2.imwrite('seam_pixel_blended_img.png', blended_img)
# cv2.imwrite('mask.png', mask.tar_nonoverlap)
# print(CB.seam_color_diff_lst.shape)