How to convert RGB PIL image to numpy array with 3 channels?

Question

I am loading image with the following code

image = PIL.Image.open(file_path)
image = np.array(image)

It works, but the size of array appears to be (X, X, 4), i.e. it has 4 layers. I would like normal RGB layers. Is it possible?

UPDATE

I found that just removing 4th channel is unsufficcient. The following code was required:

image = PIL.Image.open(file_path)
image.thumbnail(resample_size)
image = image.convert("RGB")
image = np.asarray(image, dtype=np.float32) / 255
image = image[:, :, :3]

Why?

The fourth layer is the alpha (i.e. transparency) channel. Are you sure you don't want that? — Warren Weckesser, Jul 06 '17 at 17:42
@WarrenWeckesser I am reading JPEGs, as far as I know they don't contain alpha. — Dims, Jul 06 '17 at 18:47
When I do `np.asarray(my_pil_img).shape` it returns `(480, 640, 3)` on a 480x640 .png image without alpha. Is this something that has been "fixed" in a later version since this post? — Thomas Fauskanger, Jan 25 '18 at 00:19

keredson · Answer 1 · 2017-07-06T18:52:40.870

39

The fourth layer is the transparency value for image formats that support transparency, like PNG. If you remove the 4th value it'll be a correct RGB image without transparency.

EDIT:

Example:

>>> import PIL.Image
>>> image = PIL.Image.open('../test.png')
>>> import numpy as np
>>> image = np.array(image)
>>> image.shape
(381, 538, 4)
>>> image[...,:3].shape
(381, 538, 3)

edited Jul 06 '17 at 18:52

answered Jul 06 '17 at 17:45

keredson

3,019
1
17
20

This changes shape `(64, 64, 4)` to `(22, 64, 4)` – Dims Jul 06 '17 at 18:45
Fixed the edit thanks to @Divakar's correct example. – keredson Jul 06 '17 at 18:53

score 4 · Answer 2 · answered May 26 '22 at 12:27

4

As mentioned by other answers, some images are saved with a 4th channel. To load image with just RGB channels without using numpy at all:

from PIL import Image
image = Image.open('../test.png').convert('RGB')

answered May 26 '22 at 12:27

YScharf

1,638
15
20

How to convert RGB PIL image to numpy array with 3 channels?

2 Answers2

Linked