Function transform

computer_vision/haralick_descriptors.py:104–157

Simple image transformation using one of two available filter functions: Erosion and Dilation. Args: image: binarized input image, onto which to apply transformation kind: Can be either 'erosion', in which case the :func:np.max function is called, or 'dila

(
    image: np.ndarray, kind: str, kernel: np.ndarray | None = None
)

def transform(
    image: np.ndarray, kind: str, kernel: np.ndarray | None = None
) -> np.ndarray:
    """
    Simple image transformation using one of two available filter functions:
    Erosion and Dilation.

    Args:
        image: binarized input image, onto which to apply transformation
        kind: Can be either 'erosion', in which case the :func:np.max
              function is called, or 'dilation', when :func:np.min is used instead.
        kernel: n x n kernel with shape < :attr:image.shape,
              to be used when applying convolution to original image

    Returns:
        returns a numpy array with same shape as input image,
        corresponding to applied binary transformation.

    Examples:
        >>> img = np.array([[1, 0.5], [0.2, 0.7]])
        >>> img = binarize(img, threshold=0.5)
        >>> transform(img, 'erosion')
        array([[1, 1],
               [1, 1]], dtype=uint8)
        >>> transform(img, 'dilation')
        array([[0, 0],
               [0, 0]], dtype=uint8)
    """
    if kernel is None:
        kernel = np.ones((3, 3))

    if kind == "erosion":
        constant = 1
        apply = np.max
    else:
        constant = 0
        apply = np.min

    center_x, center_y = (x // 2 for x in kernel.shape)

    # Use padded image when applying convolution
    # to not go out of bounds of the original the image
    transformed = np.zeros(image.shape, dtype=np.uint8)
    padded = np.pad(image, 1, "constant", constant_values=constant)

    for x in range(center_x, padded.shape[0] - center_x):
        for y in range(center_y, padded.shape[1] - center_y):
            center = padded[
                x - center_x : x + center_x + 1, y - center_y : y + center_y + 1
            ]
            # Apply transformation method to the centered section of the image
            transformed[x - center_x, y - center_y] = apply(center[kernel == 1])

    return transformed


def opening_filter(image: np.ndarray, kernel: np.ndarray | None = None) -> np.ndarray: