def clearn(datas):
# arrat of (tipe, warna, image)
dtipe = {}
dcolor = {}
for image in datas:
img = image[2]
img = futil.image_cleanup(img, 20)
shape = fshape.feature_shape_extract(img)
color = fcolor.feature_color_extract(img)
texture = ftexture.feature_texture_extract(img)
if not image[0] in dtipe:
dtipe[image[0]] = []
if not image[1] in dcolor:
dcolor[image[1]] = []
data = shape[0]
data.extend(shape[1:])
data.extend(texture)
dtipe[image[0]].append(data)
dcolor[image[1]].append(color)
with open('data_type.json', 'w') as ft, open('data_color.json', 'w') as fc:
json.dump(dtipe, ft)
json.dump(dcolor, fc)
return 'data_type.json', 'data_color.json'
traintype = {}
traincolor = {}
with open('data_type.json') as ft, open('data_color.json') as fc:
traintype = json.load(ft)
traincolor = json.load(fc)
for item in items:
img = Image.open(item)
img = futil.image_cleanup(img, 20)
container = {}
imgs = futil.image_clip(img)
for obj in imgs:
shape = fshape.feature_shape_extract(obj)
color = fcolor.feature_color_extract(obj)
texture = ftexture.feature_texture_extract(obj)
# Get Color
colorval = {}
colorsum = []
for key, val in traincolor.iteritems():
colorval[key] = []
for feature in val:
mean = (feature[1][0][0] - color[1][0][0]) ** 2 + (feature[1][0][1] - color[1][0][1]) ** 2 + (feature[1][0][2] - color[1][0][2]) ** 2
std = (feature[1][1][0] - color[1][1][0]) ** 2 + (feature[1][1][1] - color[1][1][1]) ** 2 + (feature[1][1][2] - color[1][1][2]) ** 2
skew = (feature[1][2][0] - color[1][2][0]) ** 2 + (feature[1][2][1] - color[1][2][1]) ** 2 + (feature[1][2][2] - color[1][2][2]) ** 2
colorval[key].append(mean + std + skew)
colorval[key].sort()
dist = sum(colorval[key][:3])
traintype = {}
traincolor = {}
with open('data_type.json') as ft, open('data_color.json') as fc:
traintype = json.load(ft)
traincolor = json.load(fc)
for item in items:
img = Image.open(item)
img = futil.image_cleanup(img, 20)
container = {}
shape = fshape.feature_shape_extract(img)
color = fcolor.feature_color_extract(img)
texture = ftexture.feature_texture_extract(img)
# Get Color
colorval = {}
colorsum = []
for key, val in traincolor.iteritems():
colorval[key] = []
for feature in val:
mean = (feature[1][0][0] - color[1][0][0]) ** 2 + (feature[1][0][1] - color[1][0][1]) ** 2 + (feature[1][0][2] - color[1][0][2]) ** 2
std = (feature[1][1][0] - color[1][1][0]) ** 2 + (feature[1][1][1] - color[1][1][1]) ** 2 + (feature[1][1][2] - color[1][1][2]) ** 2
skew = (feature[1][2][0] - color[1][2][0]) ** 2 + (feature[1][2][1] - color[1][2][1]) ** 2 + (feature[1][2][2] - color[1][2][2]) ** 2
colorval[key].append(mean + std + skew)
colorval[key].sort()
dist = sum(colorval[key][:3])
def ctest(dtype, dcolor, datas):
traintype = dtype
traincolor = dcolor
for item in datas:
img = futil.image_cleanup(item[0], 20)
shape = fshape.feature_shape_extract(img)
color = fcolor.feature_color_extract(img)
texture = ftexture.feature_texture_extract(img)
# Get Color
colorval = {}
colorsum = []
for key, val in traincolor.iteritems():
colorval[key] = []
for feature in val:
mean = (feature[1][0][0] - color[1][0][0]) ** 2 + (feature[1][0][1] - color[1][0][1]) ** 2 + (feature[1][0][2] - color[1][0][2]) ** 2
std = (feature[1][1][0] - color[1][1][0]) ** 2 + (feature[1][1][1] - color[1][1][1]) ** 2 + (feature[1][1][2] - color[1][1][2]) ** 2
skew = (feature[1][2][0] - color[1][2][0]) ** 2 + (feature[1][2][1] - color[1][2][1]) ** 2 + (feature[1][2][2] - color[1][2][2]) ** 2
colorval[key].append(mean + std + skew)
colorval[key].sort()
dist = sum(colorval[key][:2])
# print key, colorval[key], dist
if dist > thresholdc:
continue
colorsum.append((dist, key))
colorname = "uc"
if colorsum:
colorsum.sort()
colorname = colorsum[0][1]
# Get Type
data = shape[0]
data.extend(shape[1:])
data.extend(texture)
idx = range(len(weight_table))
typeval = {}
typesum = []
for key, val in traintype.iteritems():
typeval[key] = []
fx = []
for i in idx:
fx.append([])
# Apply weight and sum the squared difference of each feature
for feature in val:
for i in idx:
fx[i].append(feature[i])
typeval[key].append(sum([((data[i] - feature[i]) * weight_table[i]) ** 2 for i in idx]))
# Calculate shape distance with the average. If a shape have large distance, it might be an unknown shape.
avg = [weight_table[i] * np.mean(fx[i]) for i in idx]
std = [weight_table[i] * np.std(fx[i]) for i in idx]
dist = np.sqrt(sum([(data[i] * weight_table[i] - avg[i]) ** 2 / std[i] if abs(std[i]) > 1e-10 else 0. for i in idx]))
if dist > thresholdt:
# If it is unknown, skip it
continue
# Take 3 best match
typeval[key].sort()
typesum.append((sum(typeval[key][:3]), key))
typename = 'ut'
if typesum:
typesum.sort()
typename = typesum[0][1]
item[1].set(kamus[colorname]+' '+kamus[typename])