landingai.transform

Module for image transformations.

crop_rotated_rectangle(image, rect, angle)

Crop the input image based on the rotated rectangle. The rectangle is calculated based on the rotated rectangle's corners.

Parameters

img the input image to be cropped rect the unrotated rectangle (in parallel with the edges of the image) angle the angle of the rotated rectangle Returns

---

Tuple[np.ndarray, List[Tuple[int, int]]] the cropped image and the coordinates of the rotated rectangle

Source code in landingai/transform.py

def crop_rotated_rectangle(
    image: PIL.Image.Image,
    rect: List[Tuple[int, int]],
    angle: float,
) -> Tuple[np.ndarray, List[Tuple[int, int]]]:
    """Crop the input image based on the rotated rectangle.
    The rectangle is calculated based on the rotated rectangle's corners.

    Parameters
    ----------
    img
       the input image to be cropped
    rect
        the unrotated rectangle (in parallel with the edges of the image)
    angle
        the angle of the rotated rectangle
    Returns
    -------
    Tuple[np.ndarray, List[Tuple[int, int]]]
        the cropped image and the coordinates of the rotated rectangle
    """
    # rot_matrix = cv2.getRotationMatrix2D(center=center, angle=angle, scale=1)
    # corners = cv2.transform(np.array(rect)[None], rot_matrix)
    # quad_box = corners[0].tolist()
    # return get_minarea_rect_crop(img, corners), quad_box
    [[left, top], [right, top], [right, bottom], [left, bottom]] = rect
    center = ((left + right) / 2, (top + bottom) / 2)
    img_np = np.asarray(image)
    shape = (img_np.shape[1], img_np.shape[0])

    matrix = cv2.getRotationMatrix2D(center=center, angle=angle, scale=1)
    img_np = cv2.warpAffine(src=img_np, M=matrix, dsize=shape)
    width, height = rect[1][0] - rect[0][0], rect[3][1] - rect[0][1]
    x, y = int(center[0] - width / 2), int(center[1] - height / 2)

    img_np = img_np[y : y + height, x : x + width]
    corners = cv2.transform(np.array(rect)[None], matrix)
    quad_box: List[Tuple[int, int]] = corners[0].tolist()

    return img_np, quad_box