Class AdjustContrast

monai/transforms/intensity/array.py:1184–1244

Changes image intensity with gamma transform. Each pixel/voxel intensity is updated as:: x = ((x - min) / intensity_range) ^ gamma * intensity_range + min Args: gamma: gamma value to adjust the contrast as function. invert_image: whether to invert the image before

class AdjustContrast(Transform):
    """
    Changes image intensity with gamma transform. Each pixel/voxel intensity is updated as::

        x = ((x - min) / intensity_range) ^ gamma * intensity_range + min

    Args:
        gamma: gamma value to adjust the contrast as function.
        invert_image: whether to invert the image before applying gamma augmentation. If True, multiply all intensity
            values with -1 before the gamma transform and again after the gamma transform. This behaviour is mimicked
            from `nnU-Net <https://www.nature.com/articles/s41592-020-01008-z>`_, specifically `this
            <https://github.com/MIC-DKFZ/batchgenerators/blob/7fb802b28b045b21346b197735d64f12fbb070aa/batchgenerators/augmentations/color_augmentations.py#L107>`_
            function.
        retain_stats: if True, applies a scaling factor and an offset to all intensity values after gamma transform to
            ensure that the output intensity distribution has the same mean and standard deviation as the intensity
            distribution of the input. This behaviour is mimicked from `nnU-Net
            <https://www.nature.com/articles/s41592-020-01008-z>`_, specifically `this
            <https://github.com/MIC-DKFZ/batchgenerators/blob/7fb802b28b045b21346b197735d64f12fbb070aa/batchgenerators/augmentations/color_augmentations.py#L107>`_
            function.
    """

    backend = [TransformBackends.TORCH, TransformBackends.NUMPY]

    def __init__(self, gamma: float, invert_image: bool = False, retain_stats: bool = False) -> None:
        if not isinstance(gamma, (int, float)):
            raise ValueError(f"gamma must be a float or int number, got {type(gamma)} {gamma}.")
        self.gamma = gamma
        self.invert_image = invert_image
        self.retain_stats = retain_stats

    def __call__(self, img: NdarrayOrTensor, gamma=None) -> NdarrayOrTensor:
        """
        Apply the transform to `img`.
        gamma: gamma value to adjust the contrast as function.
        """
        img = convert_to_tensor(img, track_meta=get_track_meta())
        gamma = gamma if gamma is not None else self.gamma

        if self.invert_image:
            img = -img

        if self.retain_stats:
            mn = img.mean()
            sd = img.std()

        epsilon = 1e-7
        img_min = img.min()
        img_range = img.max() - img_min
        ret: NdarrayOrTensor = ((img - img_min) / float(img_range + epsilon)) ** gamma * img_range + img_min

        if self.retain_stats:
            # zero mean and normalize
            ret = ret - ret.mean()
            ret = ret / (ret.std() + 1e-8)
            # restore old mean and standard deviation
            ret = sd * ret + mn

        if self.invert_image: