Question

So far I’ve come up with following:

screen_array = pygame.surfarray.pixels2d(self.screen)
noise_small = numpy.random.random((self.width/4,self.height/4)) * 0.2 + 0.4
noise_big = noise_small .repeat(4, 0).repeat(4, 1)
screen_array *= noise_big

But this adds separate noise on every channel.

I know I could also use pygame.surfarray.pixels3d and then numpy.tile the noise array, but that’s way too slow for me (well, what I’m doing now is also pretty slow, but nevermind that).

Answer 1

Each element in the array returned by pygame.surfarray.pixels2d is a number obtained by packing the 3 channels into a single integer, and I don’t know of a simple method to take advantage of the array structure to do a fast and simple multiplication with this.

However, I still think your best bet is to use pygame.surfarray.pixels3d, but not to use numpy.tile. You could do something like this :

    screen_array = pygame.surfarray.pixels3d(self.screen)
    noise_small = numpy.random.random((self.width/4,self.height/4)) * 0.2 + 0.4
    noise_big = noise_small .repeat(4, 0).repeat(4, 1)
    screen_array *= noise_big[:, :, numpy.newaxis]

Notice the [:, :, numpy.newaxis]. This reshape the noise_big to be of shape (X, Y, 1) instead of (X, Y), which allows direct multiplication with the screen_array object, having shape (X, Y, 3). When that happens, the single element of the third axis is multiplied with all 3 channels. This way of tackling the problem is more efficient than using numpy.tile.