Loading Image from webpage link directly into NumPy Array (Python)

Question

I'm trying to get a JPEG image resource from the web into a NumPy array image representation similar to the array returned by scipy.misc.imread. Instead of saving the image to disk, as in the example below:

import requests
from scipy import misc
def load_image(url):
    res = requests.get(url) 
    if res == 200 and 'jpeg' in res.headers['content-type']: 
        with open('image.jpg', 'wb') as fp: 
            for chunk in res: 
                fp.write(chunk)
        img_arr = misc.imread('image.jpg') 
        return img_arr
    else: 
        return None

I'd like to load the image directly into memory. Is there a way to do so?

score 6 · Accepted Answer · answered May 18 '17 at 22:32

6

Since you mentioned scipy.misc.imread, we could use it to hide the part of Image.open. Thus, the implementation would look like this -

from scipy import misc

res = requests.get(url)
img_arr = misc.imread(BytesIO(res.content))

Performance-wise it seems comparable to the four stages of conversion as listed in the other post.

answered May 18 '17 at 22:32

Divakar

218,885
19
262
358

`misc.imread` takes a string that is a file name, not a byte stream – Nikhil Shinday May 19 '17 at 13:07
@NikhilShinday So, the posted solution didn't work for you? Worked fine for me. What's your Scipy version? – Divakar May 19 '17 at 14:35
2

module 'scipy.misc' has no attribute 'imread' – ABCD Apr 26 '21 at 00:05
Looks like they removed `imread` from scipy. There are alternatives here - https://stackoverflow.com/a/47865626/10557583 – Callum Nov 29 '21 at 12:05

score 4 · Answer 2 · answered May 18 '17 at 20:34

I found a solution that circumvents writing to disk:

from io import BytesIO
import requests
import numpy as np 
from PIL import Image
def load_image(url): 
    res = requests.get(url)
    if res == 200 and 'jpeg' in res.headers['content-type']:
        img_arr = np.array(Image.open(BytesIO(res.content)))
        return img_arr
    else: 
        return None

As far as I can tell, I'm converting between three different representations: bytes -> BytesIO -> PIL.Image -> np.array

is there a more performant way to do it?

score 0 · Answer 3 · answered Aug 23 '23 at 05:55

Since scipy.misc.imread() is deprecated, a better way would be to use cv2:

import cv2,requests
import numpy as np
from fake_useragent import UserAgent

ua = UserAgent()
img_data = requests.get(YOUR_URL_HERE,verify=False,timeout=1,
        headers={"Content-Type":ua.random}).content
img = cv2.imdecode(np.frombuffer(img_data,np.uint8),-1)
img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)

Loading Image from webpage link directly into NumPy Array (Python)

3 Answers3

Linked