How to Embed and Extract Blind Watermarks in Images with Python

Overview

Images are frequently stolen, altered, or used without permission. Traditional visible watermarks (text or logos) are easy to see and remove, while blind watermarks embed invisible data directly into the image’s frequency domain, remaining robust against common attacks such as rotation, cropping, scaling, or noise addition, and they do not require the original image for extraction.

Installation

pip install blind-watermark

Or install the latest development version:

git clone [email protected]:guofei9987/blind_watermark.git
cd blind_watermark
pip install .

Basic Usage

1. Embed a Text Watermark

Assume you have an original image ori_img.jpg and want to embed the string "开源技术小栈" as a watermark.

from blind_watermark import WaterMark

# Initialize the watermark object with image and watermark passwords (numeric keys)
wm = WaterMark(password_img=1, password_wm=1)

# Load the original image
wm.read_img('ori_img.jpg')

# Load the watermark string
wm.read_wm('开源技术小栈', mode='str')

# Embed the watermark and save the result
wm.embed('embedded.png')

print('Watermark bit length:', len(wm.wm_bit))  # prints the length of the binary watermark

After running, embedded.png looks identical to the original image, but the watermark is silently embedded.

2. Extract the Text Watermark

Extraction does not require the original image; only the watermark’s bit length and the passwords are needed.

wm_extract = WaterMark(password_img=1, password_wm=1)
wm_extract.extract('embedded.png', wm_shape=len(wm.wm_bit), mode='str')
print('Extracted watermark:', wm_extract.wm)  # outputs: 开源技术小栈

3. Embed an Image Watermark

If you want to use a small icon (e.g., a 128×128 PNG) as the watermark:

wm = WaterMark(password_img=1, password_wm=1)
wm.read_img('ori_img.jpg')
wm.read_wm('watermark_icon.png')  # the image is automatically converted to binary bits
wm.embed('embedded_with_icon.png')

To extract the image watermark:

wm_extract = WaterMark(password_img=1, password_wm=1)
wm_extract.extract('embedded_with_icon.png', wm_shape=(128, 128), out_wm_name='extracted_icon.png', mode='img')
# The file extracted_icon.png contains the recovered watermark image