Why does resizing images end up changing the channels as well?

Question

so I have this code in python using TensorFlow which opens an image and resizes it:

def parse_image(filename):
    labe