def byte_image2(byte_code):
img_feat = []
# Get image array
img = byte_make_image(byte_code)
spoints = lbp(img, 10, 10, ignore_zeros=False)
img_feat = spoints.tolist()
return img_feat
def byte_image2(byte_code):
img = byte_make_image(byte_code)
spoints = lbp(img, 10, 10, ignore_zeros=False)
return spoints.tolist()
from sklearn import svm
from sklearn.decomposition import PCA
# settings for LBP
radius = 2
n_points = 16
# 4116 dimensional lbp histogram
X = []
Y = []
with open(sys.argv[1]) as f_in:
for filename in f_in:
img = mahotas.imread(filename.rstrip(), as_grey = True)
img.resize(100,100)
lbp_hist = lbp(img, radius, n_points)
lbp_hist /= np.linalg.norm(lbp_hist)
X.append(lbp_hist)
if 'background' in filename:
Y.append(0)
else:
Y.append(1)
pca = PCA(n_components=40)
X = pca.fit_transform(X)
#clf = svm.SVC(kernel='linear')
from sklearn.linear_model import LogisticRegressionCV
clf = LogisticRegressionCV()
clf.fit(X, Y)
def worker(procnum, image_path, image_class, image_name):
print("Calculando features da imagem {}".format(procnum))
# image_path = "/home/CIT/bberton/Documents/unicamp/MC886/assignment2/data/train/Motorola-Droid-Maxx/(MotoMax)1.jpg"
image = None
try:
image = misc.imread(image_path)
except FileNotFoundError:
print("File {} not found.".format(image_path))
exit(0)
if image_class:
croped_image = img_as_float(crop_center(image, 512, 512))
else:
croped_image = img_as_float(image)
del image
# denoised_image = denoise_bilateral(croped_image, sigma_color=0.05, sigma_spatial=15, multichannel=True)
denoised_image = denoise_wavelet(croped_image, multichannel=True)
noisy_image = np.array(croped_image - denoised_image)
noisy_red, noisy_green, noisy_blue = np.array(noisy_image[:, :, 0]), \
np.array(noisy_image[:, :, 1]), \
np.array(noisy_image[:, :, 2])
lbp_red_features = lbp(noisy_red, 2, 10)
lbp_green_features = lbp(noisy_green, 2, 10)
lbp_blue_features = lbp(noisy_blue, 2, 10)
features = [
# red band
noisy_red.mean(),
np.var(noisy_red),
noisy_red.std(),
kurtosis(noisy_red.flatten()),
skew(noisy_red.flatten()),
# green band
noisy_green.mean(),
np.var(noisy_green),
noisy_green.std(),
kurtosis(noisy_green.flatten()),
skew(noisy_green.flatten()),
# blue band
noisy_blue.mean(),
np.var(noisy_blue),
noisy_red.std(),
kurtosis(noisy_blue.flatten()),
skew(noisy_blue.flatten())
]
for feature in lbp_red_features:
features.append(feature)
for feature in lbp_green_features:
features.append(feature)
for feature in lbp_blue_features:
features.append(feature)
# image class
if image_class:
features.append(class_dict[image_class])
if image_name:
features.append(image_name)
print("Features da imagem {} calculadas!".format(procnum))
return features
def gabor_composite_lbp3(image):
return lbp(gabor_composite_from_thetas(image), radius=3, points=16)
import numpy as np
from sklearn import svm
from sklearn.decomposition import PCA
# settings for LBP
radius = 2
n_points = 16
# 4116 dimensional lbp histogram
X = []
Y = []
with open(sys.argv[1]) as f_in:
for filename in f_in:
img = mahotas.imread(filename.rstrip(), as_grey=True)
img.resize(100, 100)
lbp_hist = lbp(img, radius, n_points)
lbp_hist /= np.linalg.norm(lbp_hist)
X.append(lbp_hist)
if 'background' in filename:
Y.append(0)
else:
Y.append(1)
pca = PCA(n_components=40)
X = pca.fit_transform(X)
#clf = svm.SVC(kernel='linear')
from sklearn.linear_model import LogisticRegressionCV
clf = LogisticRegressionCV()
clf.fit(X, Y)
def lbp_r1(image):
return lbp(image, radius=1, points=8)
def lbp_r2(image):
return lbp(image, radius=2, points=12)
def lbp_r3(image):
return lbp(image, radius=3, points=16)
def lbp_r2(image):
return lbp(image, radius=2, points=12)
def lbp_r1(image):
return lbp(image, radius=1, points=8)
def lbp_r3(image):
return lbp(image, radius=3, points=16)
def compute(self, image):
return lbp(image, radius=self.radius, points=self.points, ignore_zeros=False)
