[TOC]
数字水印，是指将特定的信息嵌入数字信号中，数字信号可能是音频、图片或是视频等。若要拷贝有数字水印的信号，所嵌入的信息也会一并被拷贝。数字水印可分为浮现式和隐藏式两种，前者是可被看见的水印（visible watermarking），其所包含的信息可在观看图片或视频时同时被看见。一般来说，浮现式的水印通常包含版权拥有者的名称或标志。右侧的示例图片便包含了浮现式水印。电视台在画面角落所放置的标志，也是浮现式水印的一种。 隐藏式的水印是以数字数据的方式加入音频、图片或视频中，但在一般的状况下无法被看见。隐藏式水印的重要应用之一是保护版权，期望能借此避免或阻止数字媒体未经授权的复制和拷贝。隐写术（Steganography）也是数字水印的一种应用，双方可利用隐藏在数字信号中的信息进行沟通。数字照片中的注释数据能记录照片拍摄的时间、使用的光圈和快门，甚至是相机的厂牌等信息，这也是数字水印的应用之一。某些文件格式可以包含这些称为“metadata”的额外信息。
安全性：水印信息应当难以篡改、难以伪造。 隐蔽性：水印对感官不可知觉，水印的嵌入不能影响被保护数据的可用性大大降低。不具备这一特性的水印，称为可见水印（Visible Watermarking）。如电视台播放信号的时候在某个角落经常嵌有它的标志。 强健性：水印能够抵御对嵌入后数据的一定操作，而不因为一些细微的操作而磨灭。包括数据的传输中产生的个别位错误，图像或视频、音频的压缩。不具备这一特性的水印，称为脆弱水印（Fragile Watermarking）。 水印容量：是指载体可以嵌入水印的信息量。
http://steghide.sourceforge.net/index.php Install: http://www.webm.in/2015/10/install-steghide-centos-6/ yum 问题处理： http://wolfword.blog.51cto.com/4892126/1306203
原理：https://www.researchgate.net/publication/267988699_Image_Watermarking_Using_3-Level_Discrete_Wavelet_Transform_DWT slide：https://www.slideshare.net/suritd/ppt1-48438386 Python Script： 特性:
- 采用 2 次小波变换；
- 支持水印格式：图像，文字；
- 支持加水印，解水印；
- 支持一张图像加多个水印（应对截图／剪切攻击）。
#!/usr/bin/env python
# -*- coding: UTF-8 -*-
import os
import re
import cv2
import time
import pywt
import argparse
import pygame
import numpy as np
import StringIO
from PIL import Image
pygame.init()
ORIGIN_RATE = 0.999
WATERMARK_RATE = 0.0015
TMP_PATH = "word2pic.png"
# word2img
def opencv_image_from_stringio(watermark_word):
# 用于设置画布大小和颜色
img = Image.new("RGB", (512, 512), (255, 255, 255))
font = pygame.font.Font("msyh.ttf", 50)
# 用于调整文字颜色和背景颜色
rtext = font.render(watermark_word, True, (100, 100, 100), (255, 255, 255))
sio = StringIO.StringIO()
pygame.image.save(rtext, sio)
sio.seek(0)
line = Image.open(sio)
# 用于调整文字在画布上的位置
img.paste(line, (200, 200))
img.save(TMP_PATH)
return cv2.imread(TMP_PATH)
def dwt2_single(img):
coeffs_1 = pywt.dwt2(img, 'haar', mode='reflect')
coeffs_2 = pywt.dwt2(coeffs_1[0], 'haar', mode='reflect')
return coeffs_1, coeffs_2
def dwt2(img1, img2):
coeffs1_1, coeffs1_2 = dwt2_single(img1)
coeffs2_1, coeffs2_2 = dwt2_single(img2)
return coeffs1_1, coeffs1_2, coeffs2_2
def idwt2(img, coeffs1_1_h, coeffs1_2_h):
cf2 = (img, coeffs1_2_h)
img = pywt.idwt2(cf2, 'haar', mode='reflect')
cf1 = (img, coeffs1_1_h)
img = pywt.idwt2(cf1, 'haar', mode='reflect')
return img
def channel_embedding(origin_image_chan, watermark_img_chan):
coeffs1_1, coeffs1_2, coeffs2_2 = dwt2(origin_image_chan, watermark_img_chan)
embedding_image = cv2.add(cv2.multiply(ORIGIN_RATE, coeffs1_2[0]), cv2.multiply(WATERMARK_RATE, coeffs2_2[0]))
embedding_image = idwt2(embedding_image, coeffs1_1[1], coeffs1_2[1])
np.clip(embedding_image, 0, 255, out=embedding_image)
embedding_image = embedding_image.astype('uint8')
return embedding_image
def get_watermark(args, flag):
if flag == "image":
return cv2.imread(args.watermark)
else:
return opencv_image_from_stringio(args.watermark_word)
def img_segment_embedding(watermark_img, origin_image):
origin_size = origin_image.shape[:2]
watermark_img = cv2.resize(watermark_img, (origin_size[1], origin_size[0]))
origin_image_r, origin_image_g, origin_image_b = cv2.split(origin_image)
watermark_img_r, watermark_img_g, watermark_img_b = cv2.split(watermark_img)
embedding_image_r = channel_embedding(origin_image_r, watermark_img_r)
embedding_image_g = channel_embedding(origin_image_g, watermark_img_g)
embedding_image_b = channel_embedding(origin_image_b, watermark_img_b)
embedding_image = cv2.merge([embedding_image_r, embedding_image_g, embedding_image_b])
return embedding_image
# 划分若干(num*num)块
def split_img_segments(image, num):
segments = []
if num <= 1:
segments.append(image)
return segments
ratio = 1.0/float(num)
height = image.shape[0]
width = image.shape[1]
pHeight = int(ratio*height)
pHeightInterval = (height-pHeight)/(num-1)
pWidth = int(ratio*width)
pWidthInterval = (width-pWidth)/(num-1)
for i in range(num):
for j in range(num):
x = pWidthInterval * i
y = pHeightInterval * j
cv2.imwrite('slice.png', image[y:y+pHeight, x:x+pWidth, :])
segments.append(image[y:y+pHeight, x:x+pWidth, :])
return segments
# 合并若干块
def merge_img_segments(segments, num, shape):
if num <= 1:
return segments[0]
ratio = 1.0/float(num)
height =shape[0]
width = shape[1]
channel = shape[2]
image = np.empty([height, width, channel], dtype=int)
pHeight = int(ratio*height)
pHeightInterval = (height-pHeight)/(num-1)
pWidth = int(ratio*width)
pWidthInterval = (width-pWidth)/(num-1)
cnt = 0
for i in range(num):
for j in range(num):
x = pWidthInterval * i
y = pHeightInterval * j
image[y:y+pHeight, x:x+pWidth, :] = segments[cnt]
cnt += 1
return image
# 加水印
def embedding(args, flag):
num = args.image_segments_num
origin_image = cv2.imread(args.origin)
watermark_img = get_watermark(args, flag)
# 划分若干块
origin_img_segments = split_img_segments(origin_image, num)
embedding_img_segments = []
for segment in origin_img_segments:
embedding_img_segments.append(img_segment_embedding(watermark_img, segment))
# 合并若干块
embedding_image = merge_img_segments(embedding_img_segments, num, origin_image.shape)
cv2.imwrite(args.embedding, embedding_image)
def channel_extracting(origin_image_chan, embedding_image_chan):
coeffs1_1, coeffs1_2, coeffs2_2 = dwt2(origin_image_chan, embedding_image_chan)
extracting_img = cv2.divide(cv2.subtract(coeffs2_2[0], cv2.multiply(ORIGIN_RATE, coeffs1_2[0])), WATERMARK_RATE)
extracting_img = idwt2(extracting_img, (None, None, None), (None, None, None))
return extracting_img
def img_segment_extracting(origin_image, embedding_image):
origin_image_r, origin_image_g, origin_image_b = cv2.split(origin_image)
embedding_image_r, embedding_image_g, embedding_image_b = cv2.split(embedding_image)
extracting_img_r = channel_extracting(origin_image_r, embedding_image_r)
extracting_img_g = channel_extracting(origin_image_g, embedding_image_g)
extracting_img_b = channel_extracting(origin_image_b, embedding_image_b)
extracting_img = cv2.merge([extracting_img_r, extracting_img_g, extracting_img_b])
return extracting_img
# 解水印
def extracting(args):
num = args.image_segments_num
embedding_image = cv2.imread(args.embedding)
origin_image = cv2.imread(args.origin)
origin_size = origin_image.shape[:2]
embedding_image = cv2.resize(embedding_image, (origin_size[1], origin_size[0]))
# 划分若干块
origin_img_segments = split_img_segments(origin_image, num)
embedding_img_segments = split_img_segments(embedding_image, num)
extracting_img_segments = []
for i in range (0, num*num):
extracting_img_segments.append(img_segment_extracting(origin_img_segments[i], embedding_img_segments[i]))
# 合并若干块
extracting_img = merge_img_segments(extracting_img_segments, num, origin_image.shape)
cv2.imwrite(args.extracting, extracting_img)
description = '\n'.join([
'Compares encode algs using the SSIM metric.',
' Example:',
' python watermark.py --opt embedding --origin origin.jpg --watermark watermark.jpg --embedding embedding.jpg'
])
parser = argparse.ArgumentParser(
prog='compare', formatter_class=argparse.RawTextHelpFormatter,
description=description)
parser.add_argument('--opt', default='embedding', help='embedding or extracting')
parser.add_argument('--origin', default='./samples/test.jpg', help='origin image file, length and width must be a multiple of 8')
parser.add_argument("--watermark", default='./samples/watermark.jpg', help='watermark image file')
parser.add_argument("--watermark_word", default='lzh3', help='watermark words')
parser.add_argument("--embedding", default='./samples/watermarked.jpg', help='embedding image file')
parser.add_argument("--image_segments_num", default=1, type=int, help="The sqrt number of image's segments, may be 1,2,4")
parser.add_argument("--extracting", default='./samples/extract.jpg', help='extracting image file')
args = parser.parse_args()
start = time.time()
if args.opt == 'embedding' :
embedding(args, "image")
elif args.opt == 'embedding_word':
embedding(args, "word")
elif args.opt == 'extracting':
extracting(args)
print (time.time() - start)
Result：
origin_img
水印为图像的场景:
加水印：
python watermark.py --opt embedding --origin origin.png --watermark watermark.png --embedding embedding.jpg
解水印：
python watermark.py --opt extracting --origin origin.png --embedding embedding.jpg --extracting extracting.jpg
watermark_img
embedding_img
extracting_img
水印为文字的场景:：
加多个水印(对抗截图／剪切攻击)：
python watermark.py --opt embedding_word --origin origin.png --watermark_word 'lzh3lzh3' --embedding embedding_word.jpg --image_segments_num 2
解水印：
python watermark.py --opt extracting --origin origin.png --embedding embedding_word.jpg --extracting extracting.jpg --image_segments_num 2
watermark_word: lzh3lzh3 (加2x2个水印)
embedding_img
extracting_img
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent