extract_features

extract_features.py

def extract_features(iDirName, cnn_model):
    # get face detector from dlib and initialize the face aligner
    detector = dlib.get_frontal_face_detector()

    # open the video file
    iFramePaths = [os.path.join(iDirName, iFramePath) for iFramePath in os.listdir(iDirName)]
    # an empty list to hold the results
    features = []

    # process each frame
    for iFramePath in iFramePaths:
        # read a frame from the dir
        frame = cv2.imread(iFramePath)
        frame_gray = cv2.imread(iFramePath, flags=cv2.COLOR_BGRA2GRAY)
        # convert to grayscale
        # frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

        # detect a valid face from the frame
        faces = detector(frame_gray, 2)

        # continue if more than one face is detected
        if len(faces) != 1:
            continue

        # get the coordinates of the bounding box
        (x, y, w, h) = rect_to_bb(faces[0])

        # crop the negative values
        if x < 0:
            w = w + x
            x = 0
        if y < 0:
            h = h + y
            y = 0

        # resize the image to 224x224
        # face = imutils.resize(frame[x:x + w, y:y + h], width=224, height=224)
        try:
            face = cv2.resize(frame[x:x + w, y:y + h], (224, 224), interpolation=cv2.INTER_AREA)
        except Exception as e:
            print(e)
            print(f'x={x},y={y},w={w},h={h}')
            print(f'The shape of the face is {frame[x:x + w, y:y + h].shape}')
            print(f'The shape of the frame is {frame_gray.shape}')
            print(f'The file is {iFramePath}')
            sys.exit(-1)

        # expand the shape of the image by one dimension
        face_in = np.expand_dims(face, axis=0)

        # convert to dtype 'float64'
        face_in = face_in.astype('float64')

        # some preprocessing required by the model
        face_in = preprocess_input(face_in, version=2)

        # feature extraction using CNN
        feature = cnn_model.predict(face_in).ravel()

        # add to the list
        features.append(feature)

    # convert to ndarray and return the results
    return np.array(features)