Method applyTo

utils/motionblur.py:323–411

[summary] Applies kernel to one of the following: 1. Path to image file 2. Pillow image object 3. (H,W,3)-shaped numpy array [description] Arguments: image {[str, Path, Image, np.ndarray]} keep_image_dim {bool} -- If true, the

(self, image, keep_image_dim: bool = False)

        raise NotImplementedError("Can't manually set kernel matrix yet")

    def applyTo(self, image, keep_image_dim: bool = False) -> Image:
        """[summary]
        Applies kernel to one of the following:

        1. Path to image file
        2. Pillow image object
        3. (H,W,3)-shaped numpy array
        [description]

        Arguments:
            image {[str, Path, Image, np.ndarray]}
            keep_image_dim {bool} -- If true, then we will
                    conserve the image dimension after blurring
                    by using "same" convolution instead of "valid"
                    convolution inside the scipy convolve function.

        Returns:
            Image -- [description]
        """
        # calculate kernel if haven't already
        self._createKernel()

        def applyToPIL(image: Image, keep_image_dim: bool = False) -> Image:
            """[summary]
            Applies the kernel to an PIL.Image instance
            [description]
            converts to RGB and applies the kernel to each
            band before recombining them.
            Arguments:
                image {Image} -- Image to convolve
                keep_image_dim {bool} -- If true, then we will
                    conserve the image dimension after blurring
                    by using "same" convolution instead of "valid"
                    convolution inside the scipy convolve function.

            Returns:
                Image -- blurred image
            """
            # convert to RGB
            image = image.convert(mode="RGB")

            conv_mode = "valid"
            if keep_image_dim:
                conv_mode = "same"

            result_bands = ()

            for band in image.split():

                # convolve each band individually with kernel
                result_band = convolve(
                    band, self.kernelMatrix, mode=conv_mode).astype("uint8")

                # collect bands
                result_bands += result_band,

            # stack bands back together
            result = np.dstack(result_bands)