F5.py
14.4 KB
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
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
"""
<p>This module implements the rather sophisticated F5 algorithm which was
invented by Andreas Westfeld.</p>
Unlike its vastly inferior predecessors, namely F3 and F4, it features matrix
encoding which makes it possible to embed a chunk of k bits within 2^k - 1
bits of the cover data and only change one bit (at most). A bit change is
done by subtracting the absolute value of the corresponding DCT coefficient.
When the embedding process begins, the parameter k is computed based on
the capacity of the cover image and the prospective embedding ratio.
With small amount of hidden data k becomes large which leads to a greater
embedding efficiency (embedded information per bit change).<br />
A permutation (initialized by a user-supplied seed) of the DCT coefficients
helps to scatter each chunk across the entire image.
F5 can be seen as meta-algorithm as it uses a coding scheme to change
as little data as possible and then applies a simpler algorithm (such as F3)
to actually embed data. That is why this module allows the user to specify
which embedding function (one of JSteg, F3, F4) should be used.
"""
import time
import math
import numpy as np
from stegotool.plugins.steganography.F4.F4 import F4
from stegotool.util.JPEGSteg import JPEGSteg
from stegotool.util.plugins import describe_annotate_convert
from stegotool.util.plugins import ident, ImagePath, FilePath, NewFilePath
class F5(JPEGSteg):
""" This module has two methods: <i>embed_raw_data</i> to embed data
with the F5 algorithm and <i>extract_raw_data</i> to extract data
which was embedded previously. """
def __init__(self, ui, core):
"""
Constructor of the F5 class.
"""
JPEGSteg.__init__(self, ui, core)
self._embed_hook = self._embed_k
self._extract_hook = self._extract_k
self._embed_fun = None
self.dct_p = None
self.seed = None
self.default_embedding = True
self.steg_ind = -1
self.excess_bits = None
# needed because k is embedded separately
self.cov_ind = -1
self.k_coeff = -1
@describe_annotate_convert((None, None, ident),
("cover image", ImagePath, str),
("hidden data", FilePath, str),
("stego image", NewFilePath, str),
("seed", int, int),
("embedding behavior",
['Default', 'F3', 'JSteg'], str))
def embed_raw_data(self, src_cover, src_hidden, tgt_stego, seed,
embed_fun):
"""<p>This method embeds arbitrary data into a cover image.
The cover image must be a JPEG.</p>
<p>Parameters:
<ol>
<li><pre>src_cover</pre>
A valid pathname to an image file which serves as cover image
(the image which the secret image is embedded into).</li>
<li><pre>src_hidden</pre>
A valid pathname to an arbitrary file that is supposed to be
embedded into the cover image.</li>
<li><pre>tgt_stego</pre>
Target pathname of the resulting stego image. You should save to
a PNG or another lossless format, because many LSBs don't survive
lossy compression.</li>
<li><pre>seed</pre>
A seed for the random number generator that is responsible scattering
the secret data within the cover image.</li>
<li><pre>param embed_fun</pre>
Specifies which embedding function should be used. Must be one of
'Default', 'F3', 'Jsteg'. If 'Default' is selected, the algorithm uses
the same behavior as Westfeld's implementation, i.e. decrementing
absolute values for n > 1 (F3) and using F4 in the special case n = 1.
Selecting F3 or JSteg results in using that scheme for all n.</li>
</ol>
</p>
"""
self.t0 = time.time()
self.seed = seed
if embed_fun == 'F3':
self._embed_fun = self._f3_embed
self.default_embedding = False
elif embed_fun == 'JSteg':
self._embed_fun = self._jsteg_embed
self.default_embedding = False
elif embed_fun == 'Default':
self._embed_fun = self._f3_embed
self.default_embedding = True
self.cov_ind = -1
JPEGSteg._post_embed_actions(self, src_cover, src_hidden, tgt_stego)
@describe_annotate_convert((None, None, ident),
("stego image", ImagePath, str),
("hidden data", NewFilePath, str),
("seed", int, int),
("embedding behavior", ['Default', 'F3/JSteg'],
str))
def extract_raw_data(self, src_steg, tgt_hidden, seed, embed_fun):
"""<p>This method extracts secret data from a stego image. It is
(obviously) the inverse operation of embed_raw_data.</p>
<p>Parameters:
<ol>
<li><pre>src_stego</pre>
A valid pathname to an image file which serves as stego image.</li>
<li><pre>tgt_hidden</pre>
A pathname denoting where the extracted data should be saved to.</li>
<li><pre>param seed</pre>
A seed for the random number generator that is responsible scattering
the secret data within the cover image.</li>
<li><pre>param embed_fun</pre>
Specifies which embedding function should be used. Must be one of
'Default', 'F3', 'JSteg'. If 'Default' is selected, the algorithm uses
the same behavior as Westfeld's implementation, i.e. decrementing
absolute values for n > 1 (F3) and using F4 in the special case n = 1.
Selecting F3 or JSteg results in using that scheme for all n.</li>
</ol></pre>
"""
self.t0 = time.time()
self.seed = seed
self.steg_ind = -1
if embed_fun == 'F3/JSteg':
self.default_embedding = False
elif embed_fun == 'Default':
self.default_embedding = True
# excess bits occur when the size of extracted data is not a multiple
# of k. if excess bits are available, they are prepended to hidden data
self.excess_bits = None
JPEGSteg._post_extract_actions(self, src_steg, tgt_hidden)
def _embed_k(self, cov_data, hid_data):
np.random.seed(self.seed)
self.dct_p = np.random.permutation(cov_data.size)
self.k_coeff = self._find_max_k(cov_data, hid_data)
self.ui.display_status('setting k = %d' % self.k_coeff)
k_split = self.lookup_tab.split_byte(self.k_coeff, 1)[-4:]
# embed k in F3-like style
for m in k_split:
success = False
while not success:
self.cov_ind += 1
while cov_data[self.dct_p[self.cov_ind]] == 0 or \
self.dct_p[self.cov_ind] % 64 == 0:
self.cov_ind += 1
if m != cov_data[self.dct_p[self.cov_ind]] & 1:
cov_data[self.dct_p[self.cov_ind]] -= \
math.copysign(1, cov_data[self.dct_p[self.cov_ind]])
success = cov_data[self.dct_p[self.cov_ind]] != 0
def _extract_k(self, steg_data):
# initializing the MT is done only once in order to retain the state
self.dct_p = np.random.seed(self.seed)
self.dct_p = np.random.permutation(self.steg_data.size)
k_split = np.zeros(4, np.uint8)
for i in xrange(k_split.size):
self.steg_ind += 1
while self.steg_data[self.dct_p[self.steg_ind]] == 0 or\
self.dct_p[self.steg_ind] % 64 == 0:
self.steg_ind += 1
k_split[i] = self.steg_data[self.dct_p[self.steg_ind]] & 1
self.k_coeff = self.lookup_tab.merge_words(tuple([0, 0, 0, 0] +
list(k_split)), 1)
def _find_max_k(self, cov_data, hid_data):
cnt = 4 # information about k take up 4 bits
# find number of DCT coefficients
update_cnt = 10000
for i, c in enumerate(cov_data):
if update_cnt == 0:
self._set_progress(
int(30 * (float(i) / float(cov_data.size))))
update_cnt = 10000
update_cnt -= 1
# pessimistic, but accurate estimation of the capacity of the image
ci = int(c)
if (not (ci is 0)) and (not ((i % 64) is 0)) \
and (not (ci is 1)) and (not (ci is -1)):
cnt += 1
hid_size = hid_data.size
cov_size = cnt
if cov_size < hid_size:
raise Exception("Cannot fit %d bits in %d DCT coefficients. \
Cover image is too small." % (hid_size, cov_size))
self.ui.display_status('DCT embedding ratio = %f' \
% (float(hid_size) / float(cov_size)))
k = 1
while True:
k += 1
n = (1 << k) - 1
num_chunks = cov_size / n
num_emb_bits = num_chunks * k
if num_emb_bits < hid_size:
return min(k - 1, 15)
# low level embedding functions
def _f3_embed(self, cov_data, ind):
cov_data[ind] -= math.copysign(1, cov_data[ind])
def _jsteg_embed(self, cov_data, ind):
m = 1 ^ (cov_data[ind] & 1)
cov_data[ind] = (cov_data[ind] & 0xffffe) | m
def _raw_embed(self, cov_data, hid_data, status_begin=0):
k = self.k_coeff
n = (1 << k) - 1
if n == 1 and self.default_embedding:
# in case k = n = 1, Westfeld's implementation uses F4 for
# embedding. Therefore, if 'default' embedding has been selected
# we will do the same
f4 = F4(self.ui, self.core)
f4.seed = self.seed
f4.dct_p = self.dct_p
f4.cov_ind = self.cov_ind
cov_data = f4._raw_embed(cov_data, hid_data, 30)
return cov_data
cov_ind = self.cov_ind # preventing RSI by writing 'self' less often
hid_ind = 0
remaining_bits = hid_data.size
hid_size = float(hid_data.size)
dct_p = self.dct_p
update_cnt = int(hid_size / (70.0 * k))
while remaining_bits > 0:
if update_cnt == 0:
self._set_progress(30 + int(((
hid_size - remaining_bits) / hid_size) * 70))
update_cnt = int(hid_size / (70.0 * k))
update_cnt -= 1
msg_chunk_size = min(remaining_bits, k)
msg_chunk = np.zeros(k, np.int8)
cov_chunk = np.zeros(n, np.int32)
msg_chunk[0:msg_chunk_size] = hid_data[hid_ind:hid_ind +
msg_chunk_size]
hid_ind += k
# get n DCT coefficients
for i in xrange(n):
cov_ind += 1
while cov_data[dct_p[cov_ind]] == 0 \
or dct_p[cov_ind] % 64 == 0:
cov_ind += 1
cov_chunk[i] = dct_p[cov_ind]
success = False
while not success: # loop necessary because of shrinkage
h = 0
for i in xrange(n):
h ^= ((cov_data[cov_chunk[i]] & 1) * (i + 1))
scalar_x = 0
for i in xrange(k):
scalar_x = (scalar_x << 1) + msg_chunk[i]
s = scalar_x ^ h
if s != 0:
self._embed_fun(cov_data, cov_chunk[s - 1])
else:
break
if cov_data[cov_chunk[s - 1]] == 0: # test for shrinkage
cov_chunk[s - 1:-1] = cov_chunk[s:] # adjusting
cov_ind += 1
while cov_data[dct_p[cov_ind]] == 0 or\
dct_p[cov_ind] % 64 == 0:
cov_ind += 1
cov_chunk[n - 1] = dct_p[cov_ind]
else:
success = True
remaining_bits -= k
self.k_coeff = -1 # prevent k being read from this instance
return cov_data
def _raw_extract(self, num_bits):
k = self.k_coeff
n = (1 << k) - 1
if self.is_header == None:
self.is_header = True
if n == 1 and self.default_embedding:
f4 = F4(self.ui, self.core)
f4.seed = self.seed
f4.dct_p = self.dct_p
f4.steg_data = self.steg_data
f4.is_header = self.is_header
f4.steg_ind = self.steg_ind
hid_data = f4._raw_extract(num_bits)
self.steg_ind = f4.steg_ind
self.is_header = False
return hid_data
remaining_bits = num_bits
hid_data = np.zeros(num_bits, np.uint8)
hid_ind = 0
dct_p = self.dct_p
is_header = False # signals whether or not extracting header
if self.excess_bits != None:
hid_data[hid_ind:hid_ind + self.excess_bits.size] = \
self.excess_bits
hid_ind += self.excess_bits.size
remaining_bits -= self.excess_bits.size
curr_chunk = np.zeros(k, np.uint8)
update_cnt = int(num_bits / (100.0 * k))
while remaining_bits > 0:
if update_cnt == 0 and not is_header:
self._set_progress(int(((float(num_bits) \
- remaining_bits) / num_bits) * 100))
update_cnt = int(num_bits / (100.0 * k))
update_cnt -= 1
steg_chunk = [0 for i in xrange(n)]
for i in xrange(n):
self.steg_ind += 1
while self.steg_data[dct_p[self.steg_ind]] == 0 or\
dct_p[self.steg_ind] % 64 == 0:
self.steg_ind += 1
steg_chunk[i] = self.steg_data[dct_p[self.steg_ind]]
h = 0 # hash value
for i in xrange(n):
h ^= ((steg_chunk[i] & 1) * (i + 1))
for i in xrange(k):
curr_chunk[k - i - 1] = h % 2
h /= 2
l = min(k, remaining_bits)
for i in xrange(l):
hid_data[hid_ind] = curr_chunk[i]
hid_ind += 1
# save excess bits (for later calls)
if k > remaining_bits:
self.excess_bits = curr_chunk[remaining_bits:]
else:
self.excess_bits = None
remaining_bits -= k
self.is_header = False
return hid_data
def __str__(self):
return 'F5'