def extract_lbp_features(X, P=8, R=5):
F = [] # Features are stored here
N = X.shape[0]
for k in range(N):
print("Processing image {}/{}".format(k + 1, N))
image = X[k, ...]
lbp = local_binary_pattern(image, P, R)
hist = np.histogram(lbp, bins=range(257))[0]
F.append(hist)
return np.array(F)
def extract_lbp_features(X, P=8, R=5):
"""
Extract LBP features from all input samples.
- R is radius parameter
- P is the number of angles for LBP
"""
F = [] # Features are stored here
N = X.shape[0]
for k in range(N):
print("Processing image {}/{}".format(k + 1, N))
image = X[k, ...]
lbp = local_binary_pattern(image, P, R)
hist = np.histogram(lbp, bins=range(257))[0]
F.append(hist)
return np.array(F)
class_1_path = './GTSRB_subset/class1/'
class_2_path = './GTSRB_subset/class2/'
# Load the images
for filepath in os.listdir(class_1_path):
images.append(mpimg.imread(class_1_path + filepath))
labels.append(0)
for filepath in os.listdir(class_1_path):
images.append(mpimg.imread(class_1_path + filepath))
labels.append(1)
# Extract features
for image in images:
features.append(local_binary_pattern(image))
features = np.array(features)
features = features.reshape(features.shape[0],
features.shape[1] * features.shape[2])
X_train, X_test, y_train, y_test = train_test_split(features, labels)
## Training with different classifiers
# Nearest neighbor
neigh = KNeighborsClassifier(n_neighbors=3)
scores_neigh = cross_val_score(neigh, features, labels, cv=5)
print("Neigh score:")
print(scores_neigh)
neigh.fit(X_train, y_train)
neigh_score = accuracy_score(y_test, neigh.predict(X_test))
R = 5
# histograms and corresponding classes
X = []
y = []
# class folders
class_folders = sorted(glob.glob('GTSRB_subset/*'))
for i,folder in enumerate(class_folders):
# images in class folder
name_list = glob.glob(folder+'/*')
for name in name_list:
image = plt.imread(name)
# histogram of lbp
lbp = local_binary_pattern(image, P, R)
hist = np.histogram(lbp, bins=range(257))[0]
X.append(hist)
# corresponding class
y.append(i)
# convert to numpy
X = np.array(X)
print(X.shape)
y = np.array(y)
'''
Question 5
'''
# split train and test
def extract_features(img):
lbp1 = local_binary_pattern(img, 8, 5)
feature = np.histogram(lbp1, bins=range(257))[0]
feature = np.reshape(feature, [1, 256])
return feature