def recognize_face(sess, pnet, rnet, onet, feature_array):
# Get input and output tensors
images_placeholder = sess.graph.get_tensor_by_name("input:0")
images_placeholder = tf.image.resize_images(images_placeholder, (160, 160))
embeddings = sess.graph.get_tensor_by_name("embeddings:0")
phase_train_placeholder = sess.graph.get_tensor_by_name("phase_train:0")
image_size = args.image_size
embedding_size = embeddings.get_shape()[1]
cap = cv2.VideoCapture(-1)
while (True):
ret, frame = cap.read()
gray = cv2.cvtColor(frame, 0)
if cv2.waitKey(1) & 0xFF == ord('q'):
cap.release()
cv2.destroyAllWindows()
break
if (gray.size > 0):
print(gray.size)
response, faces, bboxs = align_face(gray, pnet, rnet, onet)
print(response)
if (response == True):
for i, image in enumerate(faces):
bb = bboxs[i]
images = load_img(image, False, False, image_size)
feed_dict = {
images_placeholder: images,
phase_train_placeholder: False
}
feature_vector = sess.run(embeddings, feed_dict=feed_dict)
result, accuracy = identify_person(feature_vector,
feature_array, 8)
print(result.split("/")[2])
print(accuracy)
if accuracy < 0.9:
cv2.rectangle(gray, (bb[0], bb[1]), (bb[2], bb[3]),
(255, 255, 255), 2)
W = int(bb[2] - bb[0]) // 2
H = int(bb[3] - bb[1]) // 2
cv2.putText(gray, "Hello " + result.split("/")[2],
(bb[0] + W - (W // 2), bb[1] - 7),
cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(255, 255, 255), 1, cv2.LINE_AA)
else:
cv2.rectangle(gray, (bb[0], bb[1]), (bb[2], bb[3]),
(255, 255, 255), 2)
W = int(bb[2] - bb[0]) // 2
H = int(bb[3] - bb[1]) // 2
cv2.putText(gray, "WHO ARE YOU ?",
(bb[0] + W - (W // 2), bb[1] - 7),
cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(255, 255, 255), 1, cv2.LINE_AA)
del feature_vector
cv2.imshow('img', gray)
else:
continue
def recognize_async(images_placeholder, phase_train_placeholder, embeddings,
sess, feature_array, image, faces, bboxes):
#response, faces,bboxs = align_face(image, pnet, rnet, onet)
#response, faces, bboxs = align_face_roi(image, detector)
result_list = []
for i, image in enumerate(faces):
bb = bboxes[i]
images = load_img(image, False, False, args.image_size)
feed_dict = {
images_placeholder: images,
phase_train_placeholder: False
}
feature_vector = sess.run(embeddings, feed_dict=feed_dict)
result, accuracy = identify_person(feature_vector, feature_array, 8)
result_name = result.split("/")[-2] #.encode('utf-8')
# append result
result_list.append({
'box': bb,
'name': result_name,
'acc': accuracy,
'faceimg': image
})
del feature_vector
return result_list
def forward_pass(img, session, images_placeholder, phase_train_placeholder,
embeddings, image_size):
"""Feeds an image to the FaceNet model and returns a 128-dimension embedding for facial recognition.
Args:
img: image file (numpy array).
session: The active Tensorflow session.
images_placeholder: placeholder of the 'input:0' tensor of the pre-trained FaceNet model graph.
phase_train_placeholder: placeholder of the 'phase_train:0' tensor of the pre-trained FaceNet model graph.
embeddings: placeholder of the 'embeddings:0' tensor from the pre-trained FaceNet model graph.
image_size: (int) required square image size.
Returns:
embedding: (numpy array) of 128 values after the image is fed to the FaceNet model.
"""
# If there is a human face
if img is not None:
# Normalize the pixel values of the image for noise reduction for better accuracy and resize to desired size
image = load_img(img=img,
do_random_crop=False,
do_random_flip=False,
do_prewhiten=True,
image_size=image_size)
# Run forward pass on FaceNet model to calculate embedding
feed_dict = {images_placeholder: image, phase_train_placeholder: False}
embedding = session.run(embeddings, feed_dict=feed_dict)
return embedding
else:
return None
def recognize_face(sess, pnet, rnet, onet, feature_array):
# Get input and output tensors
images_placeholder = sess.graph.get_tensor_by_name("input:0")
images_placeholder = tf.image.resize_images(images_placeholder, (160, 160))
embeddings = sess.graph.get_tensor_by_name("embeddings:0")
phase_train_placeholder = sess.graph.get_tensor_by_name("phase_train:0")
image_size = args.image_size
embedding_size = embeddings.get_shape()[1]
#print("embedding_size:" + embedding_size)
cap = cv2.VideoCapture(0)
font = ImageFont.truetype('/System/Library/Fonts/PingFang.ttc', 18)
while (True):
ret, frame = cap.read()
gray = cv2.cvtColor(frame, 0)
if cv2.waitKey(1) & 0xFF == ord('q'):
cap.release()
cv2.destroyAllWindows()
break
if (gray.size > 0):
#print(gray.size)
print("shape: " + str(gray.shape[0]) + ", " + str(gray.shape[1]))
downsampleShape = (int(gray.shape[1] / 2), int(gray.shape[0] / 2))
gray = np.asarray(
Image.fromarray(gray).resize(downsampleShape,
resample=Image.BILINEAR))
response, faces, bboxes = align_face(gray, pnet, rnet, onet)
print(response)
if (response == True):
for i, image in enumerate(faces):
bb = bboxes[i]
images = load_img(image, False, False, image_size)
feed_dict = {
images_placeholder: images,
phase_train_placeholder: False
}
feature_vector = sess.run(embeddings, feed_dict=feed_dict)
result, accuracy = identify_person(feature_vector,
feature_array, 8)
result_name = result.split("/")[-2] #.encode('utf-8')
print(result_name)
print(accuracy)
if accuracy < 0.9:
cv2.rectangle(gray, (bb[0], bb[1]), (bb[2], bb[3]),
(255, 255, 255), 2)
W = int(bb[2] - bb[0]) // 2
H = int(bb[3] - bb[1]) // 2
#cv2.putText(gray,"Hello "+result_name,(bb[0]+W-(W//2),bb[1]-7), cv2.FONT_HERSHEY_SIMPLEX,0.5,(255,255,255),1,cv2.LINE_AA)
gray_img = Image.fromarray(gray)
draw = ImageDraw.Draw(gray_img)
draw.text((bb[0] + W - (W // 2), bb[1] - 28),
result_name,
font=font,
fill='white')
gray = np.array(gray_img)
else:
cv2.rectangle(gray, (bb[0], bb[1]), (bb[2], bb[3]),
(255, 255, 255), 2)
W = int(bb[2] - bb[0]) // 2
H = int(bb[3] - bb[1]) // 2
cv2.putText(gray, "WHO ARE YOU ?",
(bb[0] + W - (W // 2), bb[1] - 7),
cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(255, 255, 255), 1, cv2.LINE_AA)
del feature_vector
cv2.imshow('img', gray)
else:
continue