def get_cropped_face_using_fixed_eye_pos(
image_path, align_path, offset_pct, dest_size):
"""
Get face cropped and aligned to eyes from image file
Eye positions are known and corresponds to intersection in grid
:type image_path: string
:param image_path: path of image to be cropped
:type align_path: string
:param align_path: path of directory where aligned faces are saved
:type offset_pct: 2-element tuple
:param offset_pct: offset given as percentage of eye-to-eye distance
:type dest_size: 2-element tuple
:param dest_size: size of result
:rtype: OpenCV image or None
:returns: face
"""
cropped_image = None
# Open image
file_check = os.path.isfile(image_path)
if not file_check:
print('File does not exist')
return
try:
# Open whole image as PIL Image
img = Image.open(image_path)
# Align face image
(width, height) = img.size
eye_left = (width / c.GRID_CELLS_X, height / c.GRID_CELLS_Y)
eye_right = (2 * width / c.GRID_CELLS_Y, height / c.GRID_CELLS_Y)
# Create unique file path
tmp_file_name = str(uuid.uuid4()) + '.png'
tmp_file_path = os.path.join(align_path, tmp_file_name)
CropFace(
img, eye_left, eye_right, offset_pct, dest_size).save(tmp_file_path)
face_image = cv2.imread(tmp_file_path, cv2.IMREAD_GRAYSCALE)
if c.USE_HIST_EQ_IN_CROPPED_FACES:
face_image = cv2.equalizeHist(face_image)
if c.USE_NORM_IN_CROPPED_FACES:
face_image = cv2.normalize(
face_image, alpha=0, beta=255,
norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8UC1)
if c.USE_CANNY_IN_CROPPED_FACES:
face_image = cv2.Canny(face_image, 0.1, 100)
if c.USE_TAN_AND_TRIGG_NORM:
face_image = utils.normalize_illumination(face_image)
# Insert oval mask in image
if c.USE_OVAL_MASK:
face_image = utils.add_oval_mask(face_image)
return face_image
except IOError, (errno, strerror):
print "I/O error({0}): {1}".format(errno, strerror)
def get_cropped_face_from_image(
image, image_path, align_path, params, eye_cascade_classifier,
nose_cascade_classifier, face_bbox, delete_files, return_always_face):
"""
Get face cropped and aligned to eyes from image
:type image: openCV image
:param image: image to be cropped
:type image_path: string
:param image_path: path of image to be analyzed
:type align_path: string
:param align_path: path of directory where aligned faces are saved
:type params: dictionary
:param params: dictionary containing the parameters
to be used for the face detection
:type eye_cascade_classifier: CascadeClassifier
:param eye_cascade_classifier: classifier for detecting eyes
:type nose_cascade_classifier: CascadeClassifier
:param nose_cascade_classifier: classifier for detecting nose
:type face_bbox: tuple
:type face_bbox: bbox of face in original image,
represented as (x, y, width, height)
:type delete_files: boolean
:param delete_files: delete(true) or do not delete(false)
saved image files
:type return_always_face: boolean
:type return_always_face: if true,
return face even if no eyes are detected
:rtype: dictionary or None
:returns: dictionary with results
"""
result = {}
# Offset given as percentage of eye-to-eye distance
offset_pct_x = c.OFFSET_PCT_X
offset_pct_y = c.OFFSET_PCT_Y
# Final size of cropped face
cropped_face_width = c.CROPPED_FACE_WIDTH
cropped_face_height = c.CROPPED_FACE_HEIGHT
# If True, check eye positions
check_eye_positions = c.CHECK_EYE_POSITIONS
if params is not None:
if c.CROPPED_FACE_WIDTH_KEY in params:
cropped_face_width = params[c.CROPPED_FACE_WIDTH_KEY]
if c.CROPPED_FACE_HEIGHT_KEY in params:
cropped_face_height = params[c.CROPPED_FACE_HEIGHT_KEY]
if c.OFFSET_PCT_X_KEY in params:
offset_pct_x = params[c.OFFSET_PCT_X_KEY]
if c.OFFSET_PCT_Y_KEY in params:
offset_pct_y = params[c.OFFSET_PCT_Y_KEY]
if c.CHECK_EYE_POSITIONS_KEY in params:
check_eye_positions = params[c.CHECK_EYE_POSITIONS_KEY]
offset_pct = (offset_pct_x, offset_pct_y)
dest_size = (cropped_face_width, cropped_face_height)
# Detect eyes in face
eye_rects = utils.detect_eyes_in_image(image, eye_cascade_classifier)
eye_left = None
eye_right = None
eye_check_ok = False
if len(eye_rects) == 2:
x_first_eye = eye_rects[0][0]
y_first_eye = eye_rects[0][1]
w_first_eye = eye_rects[0][2]
h_first_eye = eye_rects[0][3]
x_second_eye = eye_rects[1][0]
y_second_eye = eye_rects[1][1]
w_second_eye = eye_rects[1][2]
h_second_eye = eye_rects[1][3]
x_left_eye_center = 0
y_left_eye_center = 0
x_right_eye_center = 0
y_right_eye_center = 0
if x_first_eye < x_second_eye:
x_left_eye_center = x_first_eye + w_first_eye / 2
y_left_eye_center = y_first_eye + h_first_eye / 2
x_right_eye_center = x_second_eye + w_second_eye / 2
y_right_eye_center = y_second_eye + h_second_eye / 2
else:
x_right_eye_center = x_first_eye + w_first_eye / 2
y_right_eye_center = y_first_eye + h_first_eye / 2
x_left_eye_center = x_second_eye + w_second_eye / 2
y_left_eye_center = y_second_eye + h_second_eye / 2
eye_left = (int(x_left_eye_center + face_bbox[0]),
int(y_left_eye_center + face_bbox[1]))
eye_right = (int(x_right_eye_center + face_bbox[0]),
int(y_right_eye_center + face_bbox[1]))
if check_eye_positions:
eye_check_ok = utils.check_eye_pos(
eye_left, eye_right, face_bbox, params)
else:
eye_check_ok = True
if eye_check_ok:
# Open whole image as PIL Image
img = Image.open(image_path)
# Store eye positions related to whole image
result[c.LEFT_EYE_POS_KEY] = eye_left
result[c.RIGHT_EYE_POS_KEY] = eye_right
result[c.NOSE_POSITION_KEY] = None
# Align face image
# Create unique file path
tmp_file_name = str(uuid.uuid4())
tmp_file_name_complete = tmp_file_name + '.png'
tmp_file_path = os.path.join(align_path, tmp_file_name_complete)
CropFace(
img, eye_left, eye_right, offset_pct, dest_size).save(tmp_file_path)
face_image = cv2.imread(tmp_file_path, cv2.IMREAD_GRAYSCALE)
if delete_files:
os.remove(tmp_file_path)
else:
result[c.ALIGNED_FACE_FILE_NAME_KEY] = tmp_file_name
# Check nose position
nose_check_ok = True
use_nose_pos_in_detection = c.USE_NOSE_POS_IN_DETECTION
use_nose_pos_in_recognition = c.USE_NOSE_POS_IN_RECOGNITION
if params is not None:
if c.USE_NOSE_POS_IN_DETECTION_KEY in params:
use_nose_pos_in_detection = params[c.USE_NOSE_POS_IN_DETECTION_KEY]
if c.USE_NOSE_POS_IN_RECOGNITION_KEY in params:
use_nose_pos_in_recognition = params[c.USE_NOSE_POS_IN_RECOGNITION_KEY]
if use_nose_pos_in_detection or use_nose_pos_in_recognition:
noses = utils.detect_nose_in_image(
face_image, nose_cascade_classifier)
x_right_eye = offset_pct[0] * dest_size[0]
x_left_eye = dest_size[0] - x_right_eye
y_eyes = offset_pct[1] * dest_size[1]
good_noses = 0
for(x_nose, y_nose, w_nose, h_nose) in noses:
# Coordinates of bounding box center in face image
x_center = float(x_nose + w_nose / 2)
y_center = float(y_nose + h_nose / 2)
# Store nose position relative to face image
nose_x_pct = x_center / cropped_face_width
nose_y_pct = y_center / cropped_face_height
nose = (nose_x_pct, nose_y_pct)
result[c.NOSE_POSITION_KEY] = nose
# Nose must be between eyes in horizontal direction
# and below eyes in vertical direction
if((x_center > x_right_eye) and(x_center < x_left_eye)
and (y_center > y_eyes)):
good_noses += 1
if good_noses != 1:
nose_check_ok = False
result[c.NOSE_POSITION_KEY] = None
if nose_check_ok or not use_nose_pos_in_detection:
if c.USE_HIST_EQ_IN_CROPPED_FACES:
face_image = cv2.equalizeHist(face_image)
if c.USE_NORM_IN_CROPPED_FACES:
face_image = cv2.normalize(
face_image, alpha=0, beta=255,
norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_8UC1)
if c.USE_CANNY_IN_CROPPED_FACES:
face_image = cv2.Canny(face_image, 0.1, 100)
if c.USE_TAN_AND_TRIGG_NORM:
face_image = utils.normalize_illumination(face_image)
# Insert oval mask in image
if c.USE_OVAL_MASK:
face_image = utils.add_oval_mask(face_image)
if not delete_files:
tmp_file_name_gray = (
tmp_file_name + c.ALIGNED_FACE_GRAY_SUFFIX + '.png')
tmp_file_gray_path = os.path.join(
align_path, tmp_file_name_gray)
cv2.imwrite(tmp_file_gray_path, face_image)
return result
else:
result[c.FACE_KEY] = None
return result
else:
if return_always_face:
result[c.LEFT_EYE_POS_KEY] = None
result[c.RIGHT_EYE_POS_KEY] = None
return result
else:
return None