Resizing PIL rgb image in tkinter application

background

I am creating a small slideshow application that allows for displaying an image. The original image is quite large (2000 by 2000 pixels), so I want to rescale it to fit the screen. This works as expected.

The image is essentially an array, with values ranging from 0 to 16 bit.
To get PIL to display colours, I triplicate the array, manipulate each channel individually, convert to an Image object, rescale and display. Here, the rescaling does not seem to work properly.

approach

I am sharing a barebones version of my code that repoduces this issue, at least on my system.

import tkinter as tk
import numpy as np
from PIL import Image, ImageTk

class MainApplication(tk.Frame):
    def __init__(self, parent, *args, **kwargs):
        tk.Frame.__init__(self, parent, *args, **kwargs)

        self.parent = parent

        self.images = {}

        self.load_image()
        self.show_image()


    def load_image(self):
        img = np.zeros((2000,2000))
        img[100:800, 100:800] = 255
        self.images[0] = img


    def show_image(self):

        img = self.images[0]

        img = np.repeat(img[:,:,np.newaxis], 3, axis=2)

        img = Image.fromarray(img, 'RGB')

        img = img.resize((1000,1000))
        img = ImageTk.PhotoImage(img)

        self.persistent_img = img

        self.canvas = tk.Canvas(self.parent, height=1000, width=1000)
        self.canvas.grid(row=0, column=0)
        self.canvas.create_image(0,0,anchor='nw', image=img)



if __name__ == '__main__':

    root = tk.Tk()
    MainApplication(root)
    root.mainloop()

expected

In fact, this works with the code above, if I just rescale the one-channel image:

    img = np.zeros((2000,2000))
    img[100:800, 100:800] = 255
    self.images[0] = img 
    [...]
    img = self.images[0]
    img = Image.fromarray(img)
    img = img.resize((1000,1000))
    [...]

actual

What is causing this, and is there a way around? Help is highly appreciated!

Should set dtype=uint8 in np.zeros(...) for image data:

np.zeros((2000, 2000), dtype=np.uint8)