def colordiff(A, B):
if B.shape[0] == 1:
B = np.repeat(B, A.shape[0], axis=0)
ac = np.sqrt(A[:, 1]**2 + A[:, 2]**2)
bc = np.sqrt(B[:, 1]**2 + B[:, 2]**2)
mc = (ac + bc) / 2
g = (1 - np.sqrt(mc**7 / (mc**7 + 25.0**7))) / 2.
A = A.copy()
A[:, 1] *= (1 + g)
B = B.copy()
B[:, 1] *= (1 + g)
A = color.lab2lch(A)
B = color.lab2lch(B)
dl = (B[:, 0] - A[:, 0])
dc = (B[:, 1] - A[:, 1])
dh = (B[:, 2] - A[:, 2])
mask1 = (A[:, 1] * B[:, 1] == 0)
mask2 = (~mask1) & (dh > 180)
mask3 = (~mask1) & (dh < -180)
dh[mask1] = 0
dh[mask2] -= 360
dh[mask3] += 360
dh = 2 * np.sqrt(A[:, 1] * B[:, 1]) * np.sin(np.deg2rad(dh / 2.))
ml = (A[:, 0] + B[:, 0]) / 2
mc = (A[:, 1] + B[:, 1]) / 2
mh = np.zeros_like(dh)
mh[mask1] = A[mask1, 2] + B[mask1, 2]
mh[~mask1] = mean_hue(A[~mask1, 2], B[~mask1, 2])
mls = (ml - 50)**2
sl = 1.0 + 0.015 * mls / np.sqrt(20 + mls)
# chroma weight
sc = 1 + 0.045 * mc
# hue weight
t = 1 - 0.17 * np.cos(np.deg2rad(mh - 30)) + \
0.24 * np.cos(np.deg2rad(2 * mh)) + \
0.32 * np.cos(np.deg2rad(3 * mh + 6)) - \
0.20 * np.cos(np.deg2rad(4 * mh - 63))
sh = 1 + 0.015 * mc * t
# rotation term
dtheta = 30 * np.exp(-((mh - 275) / 25)**2)
cr = 2 * np.sqrt(mc**7 / (mc**7 + 25.0**7))
tr = -np.sin(np.deg2rad(2 * dtheta)) * cr
return np.sqrt((dl/(1*sl))**2 + (dc/(1*sc))**2 + (dh/(1*sh))**2 + \
tr * (dc/(1*sc)) * (dh/(1*sh)))
def plot_segmented_colors(segmented_array, ax, colorSpace='rgb'):
"""
Counts amount of pixels of corresponding segmented colors and plots 3D diagram
Parameters
----------
segmented_array: A numpy array containing RGB color rows after color segmentation
ax : axis
colorSpace: target color space, either 'rgb' or 'hsv'
"""
seg_colors, color_counter = count_unique_rows(segmented_array)
if colorSpace == 'rgb':
if seg_colors.dtype is not 'float64':
seg_colors = util.img_as_float(seg_colors.reshape(-1, 1, 3))
plot_RGB_colors(seg_colors, ax, color_counter)
return seg_colors
elif colorSpace == 'hsv':
colors = seg_colors.copy()
seg_colors = seg_colors.reshape(seg_colors.shape[0], -1, 3)
seg_colors = skimage_color.rgb2hsv(seg_colors)
plot_HSV_colors(seg_colors, ax, colors, color_counter)
return colors, color_counter
elif colorSpace == 'LCh':
colors = seg_colors.copy()
seg_colors = seg_colors.reshape(-1, 1, 3)
seg_colors = skimage_color.lab2lch(skimage_color.rgb2lab(seg_colors))
return plot_CIELCH_colors(seg_colors, ax, colors, color_counter)
def test_rgb_lch_roundtrip(self):
rgb = img_as_float(self.img_rgb)
lab = rgb2lab(rgb)
lch = lab2lch(lab)
lab2 = lch2lab(lch)
rgb2 = lab2rgb(lab2)
assert_array_almost_equal(rgb, rgb2)
def change_hue(img, hue):
lab_img = rgb2lab(img)
lch_img = lab2lch(lab_img)
lch_img[:, :, 2] = hue
lab_img = lch2lab(lch_img)
rgb_img = lab2rgb(lab_img)
return rgb_img
def test_rgb_lch_roundtrip(self):
rgb = img_as_float(self.img_rgb)
lab = rgb2lab(rgb)
lch = lab2lch(lab)
lab2 = lch2lab(lch)
rgb2 = lab2rgb(lab2)
assert_array_almost_equal(rgb, rgb2)
def change_color_lch(batch_id, pal_lab, pal_name):
file_path = './samples/img/'
result_path = './samples/result_lch/'
file_list = os.listdir(file_path)
pal_lch = lab2lch(pal_lab)
for _, name in enumerate(file_list):
img_ori = plt.imread(file_path + name)
img_lab = rgb2lab(img_ori)
img_lch = lab2lch(img_lab)
counter_lch = [0, 0, 0, 0, 0, 0, 0]
for i in range(len(img_lch)):
for j in range(len(img_lch[i])):
point_ori = img_lch[i][j][2]
total_diff_last = 11 # lot((1-0)^2 + (1-0)^2) +1
for k in range(5):
point_pal = pal_lch[k][2]
total_diff = abs(point_ori - point_pal)
# 180 도 이상 처리 예각으로 대체
if total_diff > 5:
total_diff = 10 - total_diff
counter_lch[6] += 1
if total_diff < total_diff_last:
point_ori_tmp = point_pal
total_diff_last = total_diff
counter_fin = k
if img_lch[i][j][1] <= 0.2 and img_lch[i][j][
0] >= 9.8: # 채도 3 이하 명도 98 이상 패스(흰색)
counter_lch[5] += 1
# elif total_diff_last >= 1.2: #1.41
# counter_hsv[6] += 1
else:
counter_lch[counter_fin] += 1
img_lch[i][j][2] = point_ori_tmp
sum_total = counter_lch[0] + counter_lch[1] + counter_lch[
2] + counter_lch[3] + counter_lch[4] + counter_lch[5]
print(
f'[{pal_name}의 {name[0:-4]} 적용 내역] 팔레트1:{counter_lch[0]:,}회, 팔레트2:{counter_lch[1]:,}회, '
f'팔레트3:{counter_lch[2]:,}회, 팔레트4:{counter_lch[3]:,}회, 팔레트5:{counter_lch[4]:,}회, '
f'흰색 패스:{counter_lch[5]:,}회, 팔레트 변환 :{sum_total-counter_lch[5]:,}회'
f'({sum_total-counter_lch[5]-counter_lch[6]:,}회), 총 변환 :{sum_total:,}회'
)
img_trans = lch2lab(img_lch)
img_trans = lab2rgb(img_trans, illuminant='D55')
plt.imsave(
os.path.join(result_path,
f'{batch_id}_{pal_name}_{name[0:-4]}.png'), img_trans)
def test_lab_lch_roundtrip_dtypes(dtype):
rgb = img_as_float(data.colorwheel()).astype(dtype=dtype, copy=False)
lab = rgb2lab(rgb)
float_dtype = _supported_float_type(dtype)
assert lab.dtype == float_dtype
lab2 = lch2lab(lab2lch(lab))
decimal = 4 if float_dtype == np.float32 else 7
assert_array_almost_equal(lab2, lab, decimal=decimal)
def test_lab_lch_roundtrip(self, channel_axis):
rgb = img_as_float(self.img_rgb)
rgb = np.moveaxis(rgb, source=-1, destination=channel_axis)
lab = rgb2lab(rgb, channel_axis=channel_axis)
lab2 = lch2lab(
lab2lch(lab, channel_axis=channel_axis),
channel_axis=channel_axis,
)
assert_array_almost_equal(lab2, lab)
def rgb2lch(rgb):
"""Convert RBG to LCH colorspace (via LAB)
Input and output are in (bands, cols, rows) order
"""
# reshape for skimage (bands, cols, rows) -> (cols, rows, bands)
srgb = np.swapaxes(rgb, 0, 2)
# convert colorspace
lch = lab2lch(rgb2lab(srgb))
# return in (bands, cols, rows) order
return np.swapaxes(lch, 2, 0)
def preserve_color_cielch(content, stylized, image_name):
# extract info and convert to CIE-LCh for each image
rgbContent = io.imread(content)
labContent = color.lab2lch(
color.xyz2lab(color.rgb2xyz(numpy.asarray(rgbContent))))
labContentArray = numpy.array(labContent)
rgbStyle = io.imread(stylized)
labStyle = color.lab2lch(
color.xyz2lab(color.rgb2xyz(numpy.asarray(rgbStyle))))
labStyleArray = numpy.array(labStyle)
# color transfer
for i in range(len(labContentArray)):
for j in range(len(labContentArray[0])):
labContentArray[i][j][0] = labStyleArray[i][j][0]
labContentArray = color.xyz2rgb(
color.lab2xyz(color.lch2lab(labContentArray)))
viewer = ImageViewer(labContentArray)
viewer.show()
def get_color_map_1d(anchors):
n = len(anchors)
anchors = anchors.reshape((1, -1, 3))
x = np.mgrid[0:1:n*1j]
lab = color.rgb2lab(anchors)
lch = color.lab2lch(lab)
lch = lch.reshape((-1, 3))
L = UnivariateSpline(x, lch[:,0])
C = UnivariateSpline(x, lch[:,1])
H = UnivariateSpline(x, lch[:,2])
def curve(t):
LCH = np.stack((L(t), C(t), H(t)), axis=-1)
return color.lch2lab(LCH)
return curve
def training(request):
LAB_Info = LAB_Clothe.objects.values_list()
l = LAB_Info[LAB_Info.count()-1][1]
a = LAB_Info[LAB_Info.count()-1][2]
b = LAB_Info[LAB_Info.count()-1][3]
clothe = LAB_Info[LAB_Info.count()-1][4]
lch = color.lab2lch([float(l), float(a), float(b)])
c = lch[1]
h = lch[2] * 180 / np.pi
lab = [l, a, b]
labch = [l, a, b, c, h]
# loading data for selecting dyes
df_all = pd.read_csv(DATA_FILE_ALL)
df_all = df_all[df_all['abort'] != 1]
df_all = df_all[df_all['L'].notnull()]
df_single = pd.read_csv(DATA_FILE_SINGLE)
df_single = df_single[df_single['abort'] != 1]
""" 選擇染料組合 """
dye_selector = DyeSelector(history_num_limit=history_num_limit,
max_collection_count=max_collection_count)
possible_collections = dye_selector.get_possible_collections(df_all, df_single, labch)
""" 計算染料濃度 """
inverse_decoder = InverseDecoder(DECODER_PATH, 100)
inverse_decoder.build_graph()
with tf.Session() as sess:
inverse_decoder.init_model(sess)
pred = inverse_decoder.predict_concentrations(possible_collections, clothe, lab)
pred = modify_pred(pred)
losses = inverse_decoder.lab_loss(pred, [lab] * len(pred))
""" output result to csv """
df_output = process_output_csv(pred, losses, decoder_input_columns, clothe, clothe_type_count)
output_path = os.path.join(OUTPUT_DIR, f'{time.time()}.csv')
df_output.to_csv(output_path, index=False)
df_output = df_output[['布號', '染料_1', '濃度_1', '染料_2', '濃度_2', '染料_3', '濃度_3', '染料_4', '濃度_4', 'delta_lab']]
context = {'output_table': df_output}
return render(request, 'PigPredict_app/index.html', context)
def LabCHCalculator(frame):
imagergb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
imagelab = rgb2lab(imagergb)
imagelch = lab2lch(imagelab)
L = list()
a = list()
b = list()
C = list()
H = list()
[(L.append(y[0]), a.append(y[1]), b.append(y[2])) for x in imagelab
for y in x]
[(C.append(y[1]), H.append(y[2])) for x in imagelch for y in x]
return [
sum(L) / len(L),
sum(a) / len(a),
sum(b) / len(b),
sum(C) / len(C), (sum(H) * 57.289071045) / len(H)
]
def get_color_map_2d(anchors):
n, m, ch = anchors.shape
assert ch == 3
s, t = np.mgrid[0:1:n*1j, 0:1:m*1j]
lab = color.rgb2lab(anchors)
lch = color.lab2lch(lab)
s = s.ravel()
t = t.ravel()
h = lch[:,:,2]
for j in range(1, m):
for i in range(n):
while True:
diff = h[i, j] - h[i, j-1]
if diff > pi:
h[i, j] -= 2*pi
elif diff < -pi:
h[i, j] += 2*pi
else:
break
h_avg = np.average(h, weights=lch[:,:,1], axis=1)
reverse = np.zeros(h_avg.shape, dtype=np.uint8)
reverse[1:] = h_avg[1:] < h_avg[:-1]
reverse=np.cumsum(reverse)
print(reverse)
h += reverse.reshape((-1, 1)) * 2*pi
print(h)
l = lch[:,:,0].ravel()
c = lch[:,:,1].ravel()
h = h.ravel()
kx = min(1, n-1)
L = SmoothBivariateSpline(s, t, l, kx=kx)
C = SmoothBivariateSpline(s, t, c, kx=kx)
H = SmoothBivariateSpline(s, t, h, kx=kx)
def surface(s, t):
LCH = np.stack((L(s, t, grid=False), C(s, t, grid=False), H(s, t, grid=False)), axis=-1)
return color.lch2lab(LCH)
return surface
def componentes(c):
"""
Esta función obteniene todos los componentes, espectros y represen-
taciones de color de la image. Los formatos en los cuales se va a
transformar la imagen son: RGB, HSV, CMYK, B_LAB, LCH, R_LAB
Input:
- c: Imagen a la cual se le quieren sacar los componentes.
Output:
- (a1,a2,a3,a4,a5,a6): Posibles representaciones de la imagen.
"""
a1 = chori(c)
a2 = cvtColor(c, COLOR_BGR2HSV)
a2 = chori(a2)
a3 = phim.rgb2cmyk(np.asarray(c))
a3 = chori(a3)
a4 = cvtColor(c, COLOR_BGR2LAB)
a4_1 = chori(a4)
a5 = color.lab2lch(a4)
a5 = chori(a5)
k = a4[:, :, 2]
a6 = np.hstack((k, k, k))
return a1, a2, a3, a4_1, a5, a6
def test_lab_lch_roundtrip(self):
rgb = img_as_float(self.img_rgb)
lab = rgb2lab(rgb)
lab2 = lch2lab(lab2lch(lab))
assert_array_almost_equal(lab2, lab)
def shadow_detection(image_file, shadow_mask_file, convolve_window_size = 5, num_thresholds = 3, struc_elem_size = 5):
"""
This function is used to detect shadow - covered areas in an image, as proposed in the paper
'Near Real - Time Shadow Detection and Removal in Aerial Motion Imagery Application' by Silva G.F., Carneiro G.B.,
Doth R., Amaral L.A., de Azevedo D.F.G. (2017)
Inputs:
- image_file: Path of image to be processed for shadow removal. It is assumed that the first 3 channels are ordered as
Red, Green and Blue respectively
- shadow_mask_file: Path of shadow mask to be saved
- convolve_window_size: Size of convolutional matrix filter to be used for blurring of specthem ratio image
- num_thresholds: Number of thresholds to be used for automatic multilevel global threshold determination
- struc_elem_size: Size of disk - shaped structuring element to be used for morphological closing operation
Outputs:
- shadow_mask: Shadow mask for input image
"""
if (convolve_window_size % 2 == 0):
raise ValueError('Please make sure that convolve_window_size is an odd integer')
buffer = int((convolve_window_size - 1) / 2)
with rasterio.open(image_file) as f:
metadata = f.profile
img = rescale_intensity(np.transpose(f.read(tuple(np.arange(metadata['count']) + 1)), [1, 2, 0]), out_range = 'uint8')
img = img[:, :, 0 : 3]
lch_img = np.float32(lab2lch(rgb2lab(img)))
l_norm = rescale_intensity(lch_img[:, :, 0], out_range = (0, 1))
h_norm = rescale_intensity(lch_img[:, :, 2], out_range = (0, 1))
sr_img = (h_norm + 1) / (l_norm + 1)
log_sr_img = np.log(sr_img + 1)
del l_norm, h_norm, sr_img
gc.collect()
avg_kernel = np.ones((convolve_window_size, convolve_window_size)) / (convolve_window_size ** 2)
blurred_sr_img = cv2.filter2D(log_sr_img, ddepth = -1, kernel = avg_kernel)
del log_sr_img
gc.collect()
flattened_sr_img = blurred_sr_img.flatten().reshape((-1, 1))
labels = KMeans(n_clusters = num_thresholds + 1, max_iter = 10000).fit(flattened_sr_img).labels_
flattened_sr_img = flattened_sr_img.flatten()
df = pd.DataFrame({'sample_pixels': flattened_sr_img, 'cluster': labels})
threshold_value = df.groupby(['cluster']).min().max()[0]
df['Segmented'] = np.uint8(df['sample_pixels'] >= threshold_value)
del blurred_sr_img, flattened_sr_img, labels, threshold_value
gc.collect()
shadow_mask_initial = np.array(df['Segmented']).reshape((img.shape[0], img.shape[1]))
struc_elem = disk(struc_elem_size)
shadow_mask = np.expand_dims(np.uint8(cv2.morphologyEx(shadow_mask_initial, cv2.MORPH_CLOSE, struc_elem)), axis = 0)
del df, shadow_mask_initial, struc_elem
gc.collect()
metadata['count'] = 1
with rasterio.open(shadow_mask_file, 'w', **metadata) as dst:
dst.write(shadow_mask)
return shadow_mask
def rgb2lch(rgb):
return color.lab2lch(color.rgb2lab(rgb / 255.0))
# argparser
parser = ArgumentParser()
parser.add_argument("-l", "--l", dest="l", help="-l: 請輸入0~100的值")
parser.add_argument("-a", "--a", dest="a", help="-a: 請輸入-128~128的值")
parser.add_argument("-b", "--b", dest="b", help="-b: 請輸入-128~128的值")
parser.add_argument("-clothe",
"--clothe",
dest="clothe",
help="-clothe: 請輸入以下三種布料(FVF2184, FVF2429, FVF2666)")
args = parser.parse_args()
l, a, b, clothe = validate(format(args.l), format(args.a), format(args.b),
format(args.clothe))
# l, a, b, clothe = 29.64, -9.27, -17.33, 'FVF2666'
lch = color.lab2lch([float(l), float(a), float(b)])
c = lch[1]
h = lch[2] * 180 / np.pi
lab = [l, a, b]
labch = [l, a, b, c, h]
# loading data for selecting dyes
df_all = pd.read_csv(DATA_FILE_ALL)
df_all = df_all[df_all['abort'] != 1]
df_all = df_all[df_all['L'].notnull()]
df_single = pd.read_csv(DATA_FILE_SINGLE)
df_single = df_single[df_single['abort'] != 1]
""" 選擇染料組合 """
dye_selector = DyeSelector(history_num_limit=history_num_limit,
max_collection_count=max_collection_count)
REEXP_FOLDER_SCALE = r'\S*scale-(\d+)pc'
# ERROR:root:error: Image size (... pixels) exceeds limit of ... pixels,
# could be decompression bomb DOS attack.
# SEE: https://gitlab.mister-muffin.de/josch/img2pdf/issues/42
Image.MAX_IMAGE_PIXELS = None
#: maximal image size for visualisations, larger images will be downscaled
MAX_IMAGE_SIZE = 5000
#: define pair of forward and backward color space conversion
CONVERT_RGB = {
'rgb': (lambda img: img, lambda img: img),
'hsv': (rgb2hsv, hsv2rgb),
'lab': (rgb2lab, lab2rgb),
'luv': (rgb2luv, luv2rgb),
'hed': (rgb2hed, hed2rgb),
'lch':
(lambda img: lab2lch(rgb2lab(img)), lambda img: lab2rgb(lch2lab(img))),
}
def detect_binary_blocks(vec_bin):
""" detect the binary object by beginning, end and length in !d signal
:param list(bool) vec_bin: binary vector with 1 for an object
:return tuple(list(int),list(int),list(int)):
>>> vec = np.array([1] * 15 + [0] * 5 + [1] * 20)
>>> detect_binary_blocks(vec)
([0, 20], [15, 39], [14, 19])
"""
begins, ends, lengths = [], [], []
def test_lab_lch_3d(self):
lab0 = self._get_lab0()
lch0 = lab2lch(lab0)
lch3 = lab2lch(lab0[None, None, None, :])
assert_array_almost_equal(lch0, lch3[0, 0, 0, :])
if img.dtype != 'float32':
img = cv2.normalize(img.astype('float32'), None, 0.0, 1.0, cv2.NORM_MINMAX)
# get dimensions of input image
x, y, z = img.shape
# store a copy of original image for comparison
img1 = np.zeros((x, y, z))
img1 = img.copy()
alpha = None
# Convert to colorspaces CIELAB, CIELUV, and CIELCH
lab = color.rgb2lab(img) # - RGB to LAB
luv = color.rgb2luv(img) # - RGB to LUV
LCH = color.lab2lch(img) # - LAB to LCH
# Compute G(x, y) for image
Gx = np.zeros((x, y)) # Create empty array for Gx component
Gy = np.zeros((x, y)) # Create empty array for Gy component
for i in range(1, x-1):
for j in range(1, y-1):
#Calculate the Gx and Gy per pixel using color difference from Eq 3 (on pg 2 of paper)
Gx[i, j] = color_difference(lab[i + 1, j, :], lab[i - 1, j, :], luv[i + 1, j, :], luv[i - 1, j, :], alpha)
Gy[i, j] = color_difference(lab[i, j + 1, :], lab[i, j - 1, :], luv[i, j + 1, :], luv[i, j - 1, :], alpha)
# Assign Lightness, chroma, and hue arrays from LCH to their own respective arrays
L = LCH[:, :, 0]
C = LCH[:, :, 1]
H = LCH[:, :, 2]
T = np.zeros((x, y, 9))
def rgb2lch(rgb):
lab = color.rgb2lab(rgb)
return color.lab2lch(lab)
def lab2lch(LabPic):
color.adapt_rgb.each_channel
LchPic = color.lab2lch(LabPic)
return (LchPic)
def lightness(pixel):
return lab2lch(rgb2lab(pixel))[:, :, 0]
def test_lab_lch_roundtrip(self):
rgb = img_as_float(self.img_rgb)
lab = rgb2lab(rgb)
lab2 = lch2lab(lab2lch(lab))
assert_array_almost_equal(lab2, lab)
def chroma(pixel):
return lab2lch(rgb2lab(pixel))[:, :, 1]
def test_lab_lch_3d(self):
lab0 = self._get_lab0()
lch0 = lab2lch(lab0)
lch3 = lab2lch(lab0[None, None, None, :])
assert_array_almost_equal(lch0, lch3[0, 0, 0, :])
def hue(pixel):
return lab2lch(rgb2lab(pixel))[:, :, 2]
