How do I read an image from a path with Unicode characters?

Question

I have the following code and it fails, because it cannot read the file from disk. The image is always None.

# -*- coding: utf-8 -*-
import cv2
import numpy

bgrImage = cv2.imread(u'D:\\ö\\handschuh.jpg')

Note: my file is already saved as UTF-8 with BOM. I verified with Notepad++.

In Process Monitor, I see that Python is acccessing the file from a wrong path:

I have read about:

• Open file with unicode filename, which is about the open() function and not related to OpenCV.
• How do I read an image file using Python, but that's unrelated to Unicode issues.

Answer 1

It can be done by

• opening the file using open(), which supports Unicode as in the linked answer,
• read the contents as a byte array,
• convert the byte array to a NumPy array,
• decode the image

# -*- coding: utf-8 -*-
import cv2
import numpy

stream = open(u'D:\\ö\\handschuh.jpg', "rb")
bytes = bytearray(stream.read())
numpyarray = numpy.asarray(bytes, dtype=numpy.uint8)
bgrImage = cv2.imdecode(numpyarray, cv2.IMREAD_UNCHANGED)

Answer 2

Inspired by Thomas Weller's answer, you can also use np.fromfile() to read the image and convert it to ndarray and then use cv2.imdecode() to decode the array into a three-dimensional numpy ndarray (suppose this is a color image without alpha channel):

import numpy as np

# img is in BGR format if the underlying image is a color image
img = cv2.imdecode(np.fromfile('测试目录/test.jpg', dtype=np.uint8), cv2.IMREAD_UNCHANGED)

np.fromfile() will convert the image on disk to numpy 1-dimensional ndarray representation. cv2.imdecode can decode this format and convert to the normal 3-dimensional image representation. cv2.IMREAD_UNCHANGED is a flag for decoding. Complete list of flags can be found here.

PS. For how to write image to a path with unicode characters, see here.

Answer 3

I copied them to a temporary directory. It works fine for me.

import os
import shutil
import tempfile

import cv2


def cv_read(path, *args):
    """
    Read from a path with Unicode characters.

    :param path: path of a single image or a directory which contains images
    :param args: other args passed to cv2.imread
    :return: a single image or a list of images
    """
    with tempfile.TemporaryDirectory() as tmp_dir:
        if os.path.isdir(path):
            shutil.copytree(path, tmp_dir, dirs_exist_ok=True)
        elif os.path.isfile(path):
            shutil.copy(path, tmp_dir)
        else:
            raise FileNotFoundError

        img_arr = [
            cv2.imread(os.path.join(tmp_dir, img), *args)
            for img in os.listdir(tmp_dir)
        ]

        return img_arr if os.path.isdir(path) else img_arr[0]

Answer 4

My problem is similar to you, however, my program will terminate at the image = cv2.imread(filename)statement.

I solved this problem by first encode the file name into utf-8 and then decode it as

 image = cv2.imread(filename.encode('utf-8', 'surrogateescape').decode('utf-8', 'surrogateescape'))

Answer 5

bgrImage = cv2.imread(filename.encode('utf-8'))

encode file full path to utf-8