Previously, we introduced that OpenCV can read local images using cv2.imread(), but this method cannot directly load images from a network URL.

Reading Network Images: cv2.imdecode

To read images from a network in OpenCV, you need to use cv2.imdecode(). This method can directly process byte streams (such as images transmitted over the network or camera frames) without saving them to a file first.

Syntax:

image = cv2.imdecode(buf, flags)

• buf: Must be a numpy.ndarray, typically created by converting byte data using np.frombuffer(). The NumPy array's dtype must be np.uint8, because image encoding data is essentially a byte stream with values ranging from 0 to 255, which corresponds to 8-bit unsigned integers.
• flags: Optional parameter specifying the decoding mode (same as flags in cv2.imread). It determines the color mode, transparency channel retention, etc.

Common flags values:

Example:

import cv2
import numpy as np
import ssl
import urllib.request

if __name__ == "__main__":
    url = "https://p9-passport.byteacctimg.com/img/user-avatar/6c49bd0b908f5b1601050a168d0283b2~60x60.awebp"
    headers = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36"}
    req = urllib.request.Request(url, headers=headers)
    context = ssl._create_unverified_context()
    response = urllib.request.urlopen(req, context=context)
    img_bytes = response.read()
    img_array = np.frombuffer(img_bytes, dtype=np.uint8)
    image = cv2.imdecode(img_array, cv2.IMREAD_UNCHANGED)
    cv2.imshow("Image from URL", image)
    key = cv2.waitKey(0)
    if key == ord("q"):
        cv2.destroyAllWindows()

In this example, we send a request with a browser-like User-Agent to fetch an avatar image. The response bytes are converted to a NumPy array and decoded using cv2.imdecode. The image is displayed until the 'q' key is pressed.

Why Both cv2.imread and cv2.imdecode?

Although both functions convert encoded data into an image matrix, they follow the Single Responsibility Principle (SRP). cv2.imread handles file I/O and decoding, while cv2.imdecode focuses solely on decoding from a byte buffer. Here's a comparison:

This design provides flexibility for handling images from various sources (network, databases, memory) without requiring the overhead of file operations.