3

I have an image converted to a ndarray with RGBA values. Suppose it's 50 x 50 x 4.

I want to replace all the pixels with values array([255, 255, 255, 255]) for array([0, 0, 0, 0]). So:

from numpy import *
from PIL import Image
def test(mask):
        mask = array(mask)
        find = array([255, 255, 255, 255])
        replace = array([0, 0, 0, 0])
        return putmask(mask, mask != find, replace)

mask = Image.open('test.png')
test(mask)

What am I doing wrong? That gives me a ValueError: putmask: mask and data must be the same size. Yet if I change the arrays to numbers (find = 255, replace = 0) it works.

dolans
  • 33
  • 3

3 Answers3

4

A more concise way to do this is

img = Image.open('test.png')
a = numpy.array(img)
a[(a == 255).all(axis=-1)] = 0
img2 = Image.fromarray(a, mode='RGBA')

More generally, if the items of find and repl are not all the same, you can also do

find = [1, 2, 3, 4]
repl = [5, 6, 7, 8]
a[(a == find).all(axis=-1)] = repl
Sven Marnach
  • 574,206
  • 118
  • 941
  • 841
2

One way to do this kind of channel masking is to split the array into r,g,b,a channels, then define the index using numpy logical bit operations:

import numpy as np
import Image

def blackout(img):
    arr = np.array(img)
    r,g,b,a=arr.T
    idx = ((r==255) & (g==255) & (b==255) & (a==255)).T
    arr[idx]=0
    return arr

img = Image.open('test.png')
mask=blackout(img)
img2=Image.fromarray(mask,mode='RGBA')
img2.show()
unutbu
  • 842,883
  • 184
  • 1,785
  • 1,677
2

This solution uses putmask and I think is the closest to the OPs code. There are two errors in the original code that the OP should know about: 1) putmask is an in-place operation. It returns None. 2) putmask also requires equal-sized arrays. It (too bad) doesn't have an axis keyword argument.

import numpy as np
from PIL import Image

img1 = Image.open('test.png')
arry = np.array(img1)
find = np.array([255, 255, 255, 255])
repl = np.array([  0,   0,   0,   0])
# this is the closest to the OPs code I could come up with that
# compares each pixel array with the 'find' array
mask = np.all(arry==find, axis=2)
# I iterate here just in case repl is not always the same value
for i,rep in enumerate(repl):
    # putmask works in-place - returns None
    np.putmask(arry[:,:,i], mask, rep)

img2 = Image.fromarray(arry, mode='RGBA')
img2.save('testx.png')
Paul
  • 42,322
  • 15
  • 106
  • 123
  • 1. Why do you use `np.logical_not(imga-find)` instead of `imga != find`? 2. What is `imga` anyway? 3. How would you avoid the loop in the case that all entries in `repl` are the same? – Sven Marnach Apr 04 '11 at 09:01
  • @ Sven Marnach: 2) Thanks `imga` should have been `arry`. 1) I wasn't aware you could broadcast with `==`. 3) I'd create a 3D mask and `putmask(arry, mask, 0)`. – Paul Apr 04 '11 at 12:55