Programmatically Detect Corrupted Image

When we download a huge image dataset from the internet, occasionally we may encounter image corruption due to incomplete download. For example:

In [1]: import cv2
In [2]: img = cv2.imread('corrupted.jpg')
Premature end of JPEG file

Here we demonstrate how to programmatically capture this error before sending these corrupted images into neural network.

from skimage import io
def verify_image(img_file):
    try:
        img = io.imread(img_file)
    except:
        return False
    return True

Unlike cv2.imread, which only print error message, and fill corrupted image with grey color.

A corrupted image is filled with grey color

skimage.io will throw exception and allow us to detect and download again.

OSError: image file is truncated (3 bytes not processed)