Trying to calculate the mean of a sliding window of an image Python

Question

I'm trying to pixelate (\mosaic) an image by calculate the mean of a (non overlap) sliding window over the image. For this I try to implement a "window size" and a "step" parameters. Assuming my step won't exceed the image border. Means that if my image is a 32X32 dims the window can be 2x2\4x4\8x8\16x16 dims. Here an example

I try to look for some combinations of mean operator\mask\convolution but didn't find anything relevant.

Here some examples of what iI try to look for: Those links gave some parts of my question but iI didn't find out how to combine them in order to implement a sliding window with step skipping.

Numpy Two-Dimensional Moving Average, scipy.org/../scipy.signal.medfilt, mosaic.py on GitHub and Numpy Vectorization of sliding-window operation How to do this sliding window in order to pixelate parts of an image seperatly.

Are you looking just for downsampling (e.g. [this question](https://stackoverflow.com/questions/18666014/downsample-array-in-python)) or do you actually want to preserve the image size (something like a downsample and upscale)? — jdehesa, Sep 02 '17 at 15:10
I'm not sure If I understood correctly what downsampling means. I want to keep the dimensions in their original proportional size. Just want to move a non overlap window and apply the average value of it on the corresponding part of the image. Then to slide the window to the next part repeativley. Exectly as shown in the image I attached. Thank you for your answer. — Evya IL, Sep 02 '17 at 20:35

score 0 · Answer 1 · answered Sep 04 '17 at 11:12

Here is (I think) a possible solution to your problem:

def pixelate(img, wx, wy=None):
    wy = wy or wx
    y, x = img.shape
    if x % wx != 0 or y % wy != 0:
        raise ValueError("Invalid window size.")
    ny = y // wy
    nx = x // wx
    windowed = img.reshape((ny, wy, nx, wx))
    means = windowed.mean(axis=(1, 3), keepdims=True)
    means = np.tile(means, (1, wy, 1, wx))
    result = means.reshape((y, x))
    return result

Where img is a 2D NumPy array representing an image, wx is the horizontal size of the window and wy the vertical size (which defaults to the same as wy). The image must be divisible by the window size. Basically it reshapes the image array to its windows, computes the means, tiles the result and reshapes back.

Here is an example with a circumference:

import numpy as np
import matplotlib.pyplot as plt

def pixelate(img, wx, wy=None):
    wy = wy or wx
    y, x = img.shape
    if x % wx != 0 or y % wy != 0:
        raise ValueError("Invalid window size.")
    ny = y // wy
    nx = x // wx
    windowed = img.reshape((ny, wy, nx, wx))
    means = windowed.mean(axis=(1, 3), keepdims=True)
    means = np.tile(means, (1, wy, 1, wx))
    result = means.reshape((y, x))
    return result

# Build a circumference
WIDTH = 400
HEIGHT = 300
RADIUS = 100
THICKNESS = 10
xx, yy = np.meshgrid(np.arange(WIDTH) - WIDTH / 2, np.arange(HEIGHT) - HEIGHT / 2)
r = np.sqrt(np.square(xx) + np.square(yy))
circ = (r > (RADIUS - THICKNESS / 2)) & (r < (RADIUS + THICKNESS / 2))
circ = circ.astype(np.float32)

# Pixelate
WINDOW_SIZE = 20
circ_pix = pixelate(circ, WINDOW_SIZE)

# Show
fig = plt.figure()
ax1 = fig.add_subplot(121)
ax1.imshow(circ, "binary")
ax2 = fig.add_subplot(122)
ax2.imshow(circ_pix, "binary")

Output:

Trying to calculate the mean of a sliding window of an image Python

1 Answers1