Custom image interpolation kernels

When interpolation is set to ‘custom’, the convolution kernel provided by custom_interpolation_kernel_2d is used to convolve the image on the gpu. In this example, we use custom gaussian kernels of arbitrary size, a sharpening kernel and a ridge detection kernel.

Under the hood, this works by by sampling the image texture with linear interpolation in a regular grid (of size = of the kernel) around each fragment, and then using the weights in the kernel to add up the final fragment value.

Tags: gui, visualization-nD

import numpy as np
from magicgui import magicgui
from scipy.signal.windows import gaussian
from skimage import data

import napari

viewer = napari.view_image(data.astronaut(), rgb=True, interpolation='custom')


def gaussian_kernel(size, sigma):
    window = gaussian(size, sigma)
    kernel = np.outer(window, window)
    return kernel / kernel.sum()


def sharpen_kernel():
    return np.array([
        [ 0, -1,  0],
        [-1,  5, -1],
        [ 0, -1,  0],
    ])


def ridge_detection_kernel():
    return np.array([
        [-1, -1, -1],
        [-1,  9, -1],
        [-1, -1, -1],
    ])


@magicgui(
    auto_call=True,
    kernel_size={"widget_type": 'Slider', "min": 1, "max": 20},
    sigma={"widget_type": 'FloatSlider', "min": 0.1, "max": 5, "step": 0.1},
    kernel_type={"choices": ['none', 'gaussian', 'sharpen', 'ridge_detection']},
)
def gpu_kernel(image: napari.layers.Image, kernel_type: str = 'gaussian', kernel_size: int = 5, sigma: float = 1):
    if kernel_type == 'none':
        image.interpolation2d = 'linear'
    else:
        image.interpolation2d = 'custom'

    if kernel_type == 'gaussian':
        gpu_kernel.kernel_size.show()
        gpu_kernel.sigma.show()
    else:
        gpu_kernel.kernel_size.hide()
        gpu_kernel.sigma.hide()

    if kernel_type == 'gaussian':
        image.custom_interpolation_kernel_2d = gaussian_kernel(kernel_size, sigma)
    elif kernel_type == 'sharpen':
        image.custom_interpolation_kernel_2d = sharpen_kernel()
    elif kernel_type == 'ridge_detection':
        image.custom_interpolation_kernel_2d = ridge_detection_kernel()


viewer.window.add_dock_widget(gpu_kernel)
gpu_kernel()


if __name__ == '__main__':
    napari.run()