-
Notifications
You must be signed in to change notification settings - Fork 0
/
deep_steganography.py
232 lines (181 loc) · 8.56 KB
/
deep_steganography.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
from keras.callbacks import ModelCheckpoint, LearningRateScheduler, TensorBoard
from keras.layers import *
from keras.models import Model
from keras.preprocessing import image
import keras.backend as K
import matplotlib.pyplot as plt
import numpy as np
import os
import random
import scipy.misc
from tqdm import *
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
# %matplotlib inline
import keras
from keras.models import Model,Sequential
from keras.datasets import mnist
from tqdm import tqdm
from keras.layers.advanced_activations import LeakyReLU
from keras.optimizers import adam
import cv2
import shutil
from keras.utils import to_categorical
from keras.utils import np_utils
from keras.preprocessing.image import ImageDataGenerator
from keras.datasets import cifar10
from keras import regularizers
from keras.callbacks import LearningRateScheduler
from scipy import ndimage
from keras.callbacks import Callback,ModelCheckpoint
from keras.models import Sequential,load_model
from keras.wrappers.scikit_learn import KerasClassifier
from sklearn.preprocessing import OneHotEncoder
from __future__ import print_function, division
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.convolutional import UpSampling2D, Conv2D
from keras.optimizers import Adam
files = os.listdir('tiny-imagenet-200/train')
files_te = os.listdir('tiny-imagenet-200/test/images')
x_train = np.empty((2000,64,64,3), 'uint64')
a=0
for i in range(200):
idd = np.random.randint(0, 500, 10)
for j in range(10):
image = cv2.imread('tiny-imagenet-200/train/'+files[i]+'/images/'+files[i]+'_'+str(idd[j])+'.JPEG')
x_train[a] = image
a=a+1
x_test = np.empty((2000,64,64,3), 'uint64')
a=0
for i in range(2000):
image = cv2.imread('tiny-imagenet-200/test/images/'+files_te[i])
x_test[a] = image
a=a+1
input_S = x_train[0:1000]
input_C = x_train[1000:]
input_C = input_C/255.0
input_S = input_S/255.0
beta = 1.0
def rev_loss(true,pred):
loss = beta*K.sum(K.square(true-pred))
return loss
def full_loss(true,pred):
message_true, container_true = true[...,0:3], true[...,3:6]
message_pred, container_pred = pred[...,0:3], pred[...,3:6]
message_loss = rev_loss(message_true, message_pred)
container_loss = K.sum(K.square(container_true-container_pred))
loss = message_loss + container_loss
return loss
def prep_and_hide_network(input_size):
input_message = Input(shape=(input_size))
input_cover = Input(shape=(input_size))
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(input_message)
x2 = Conv2D(10, (4,4), strides = (1,1), padding = 'same', activation = 'relu')(input_message)
x3 = Conv2D(5, (5,5), strides = (1,1), padding = 'same', activation = 'relu')(input_message)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (4,4), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (5,5), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x = concatenate([input_cover,x])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (4,4), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (5,5), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (4,4), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (5,5), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (4,4), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (5,5), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (4,4), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (5,5), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (4,4), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (5,5), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
image_container = Conv2D(3, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
encoder = Model(inputs = [input_message, input_cover],
outputs = image_container)
return encoder
def reveal_network(input_size, fixed=False):
reveal_input = Input(shape=(input_size))
input_with_noise = GaussianNoise(0.01)(reveal_input)
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(input_with_noise)
x2 = Conv2D(10, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(input_with_noise)
x3 = Conv2D(5, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(input_with_noise)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
x1 = Conv2D(50, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x2 = Conv2D(10, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x3 = Conv2D(5, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
x = concatenate([x1, x2, x3])
message = Conv2D(3, (3,3), strides = (1,1), padding = 'same', activation = 'relu')(x)
reveal = Model(inputs = reveal_input,
outputs = message)
return reveal
shape = input_S.shape[1:]
input_message = Input(shape = (shape))
input_container = Input(shape = (shape))
prep_and_hide = prep_and_hide_network(shape)
reveal = reveal_network(shape)
reveal.compile(optimizer = 'adam',
loss = rev_loss)
reveal.trainable = False
output_container = prep_and_hide([input_message, input_container])
output_message = reveal(output_container)
deep_stean = Model(inputs = [input_message, input_container],
outputs = concatenate([output_message, output_container]))
deep_stean.compile(optimizer = 'adam',
loss = full_loss)
def lr_schedule(epoch_idx):
if epoch_idx < 200:
return 0.001
elif epoch_idx < 400:
return 0.0003
elif epoch_idx < 600:
return 0.0001
else:
return 0.00003
m = input_S.shape[0]
loss_history = []
batch_size = 32
for epoch in range(1000):
np.random.shuffle(input_S)
np.random.shuffle(input_C)
itera = int(m/batch_size)
f_loss_mean = 0
r_loss_mean = 0
for i in range(itera):
batch_message = input_S[i*batch_size:min((i+1)*batch_size,m)]
batch_cover = input_C[i*batch_size:min((i+1)*batch_size,m)]
container = prep_and_hide.predict([batch_message,batch_cover])
f_loss = deep_stean.train_on_batch(x = [batch_message,batch_cover],
y = np.concatenate((batch_message,batch_cover), axis = 3))
r_loss = reveal.train_on_batch(x = container,
y = batch_message)
f_loss_mean = f_loss_mean + f_loss
r_loss_mean = r_loss_mean + r_loss
print('Epoch = '+str(epoch)+' batch = '+str(i)+' | full loss = '+str(f_loss)+' | rev_loss = '+str(r_loss))
f_loss_mean = f_loss_mean/itera
r_loss_mean = r_loss_mean/itera
print('Epoch = '+str(epoch)+' | mean full loss = '+str(f_loss_mean)+' | mean rev_loss = '+str(r_loss_mean))
print('--------------------Epoch '+str(epoch)+' complete--------------------')
K.set_value(deep_stean.optimizer.lr, lr_schedule(epoch))
K.set_value(reveal.optimizer.lr, lr_schedule(epoch))