def predict(self, image_path, threshold=0.85):
utils = IU()
model = load_model('models/predictor_NN_model.h5')
image = cv2.imread(image_path)
locs = utils.get_face_locations(image=image,
model=self.face_model,
scaleup=self.upsample)
encodings = list()
for (sX, sY, eX, eY) in locs:
face = cv2.resize(image[sY:eY, sX:eX], (160, 160))
encoding = utils.face_encodings(face,
model=self.encoding_model,
jitters=self.jitters)
if len(encoding) == 0:
continue
else:
encodings.append(encoding[0])
if len(encodings) == 0:
return False
data = np.vstack(encodings)
predictions = model.predict(data)
for prediction in predictions:
if prediction[0] > threshold:
return True
return False
def convolve2d(img, kernel):
"""
Performs convolution in 2D array.
:param img: numpy 2D array: input array
:param kernel: array 3x3: convolution kernel
:return: numpy 2D array: result of convolution, image has same shape as input
"""
kernel = np.flip(kernel, 0)
kernel = np.flip(kernel, 1)
output = np.zeros(img.shape)
window_size = 3
padded_input = ImageUtilities.pad_image(img, 1)
h, w = img.shape
for x in range(0, w):
for y in range(0, h):
window = padded_input[y:y + window_size, x:x + window_size]
window = np.multiply(window, kernel)
output[y, x] = window.sum()
return output
def median_filter(img, window_size):
"""
Filters image using median filter.
:param img: numpy 2D array: array to filter
:param window_size: int: size of filter
:return: numpy 2D array: filtered image, same shape as input
"""
if Config.USE_CV2_FILTERS:
return cv2.medianBlur(img, 3)
output = np.zeros(img.shape)
padded_input = ImageUtilities.pad_image(img, 1)
half_win_size = window_size // 2
h, w = img.shape
mid = (window_size**2) // 2
for x in range(0, w):
for y in range(0, h):
window = padded_input[y:y + window_size, x:x + window_size]
window = window.flatten()
window.sort()
val = window[mid]
output[y][x] = val
return output
def generate_training_set(self):
if self.paths is None or len(self.paths) < 10:
error = QErrorMessage()
error.showMessage('Select atleast 10 images')
error.exec_()
else:
utils = IU()
self.status.setText('Generating training data...')
encodings = list()
for image_path in self.paths:
image = cv2.imread(image_path)
(sX, sY, eX, eY) = utils.get_face_locations(image=image)[0]
face = cv2.resize(image[sY:eY, sX:eX], (200, 200))
face = face[:, :, ::-1]
encoding = FR.face_encodings(face,
num_jitters=10,
known_face_locations=[(sX, sY, eX, eY)])
if len(encoding) == 0:
pass
else:
encodings.append(encoding[0])
with open('models/training_data.clf', 'wb') as file:
pickle.dump(encodings, file)
self.status.setText('Done!')
