8cfc1a23
Chunk
F5 half-finished.
|
1
2
|
__author__ = 'chunk'
|
c9fdeb00
Chunk
LSB and F4 added.
|
3
|
"""
|
26e2fe9f
Chunk
MPB steganalysis ...
|
4
5
|
ref - http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.115.3651&rep=rep1&type=pdf
|
c9fdeb00
Chunk
LSB and F4 added.
|
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
|
<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.
"""
|
c9fdeb00
Chunk
LSB and F4 added.
|
26
27
|
import time
import math
|
033d3b0d
Chunk
staged.
|
28
|
import numpy as np
|
b69b6985
Chunk
py module refract...
|
29
|
import numpy.random as rnd
|
033d3b0d
Chunk
staged.
|
30
|
from msteg.StegBase import *
|
c6c61f81
Chunk
staged.
|
31
|
from F4 import F4
|
033d3b0d
Chunk
staged.
|
32
|
import mjsteg
|
c9fdeb00
Chunk
LSB and F4 added.
|
33
34
|
import jpegObj
from common import *
|
033d3b0d
Chunk
staged.
|
35
|
|
c9fdeb00
Chunk
LSB and F4 added.
|
36
37
38
39
|
class F5(StegBase):
""" 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
|
8cfc1a23
Chunk
F5 half-finished.
|
40
|
which was embedded previously. """
|
c9fdeb00
Chunk
LSB and F4 added.
|
41
42
43
|
def __init__(self, key=sample_key, k=None):
"""
|
033d3b0d
Chunk
staged.
|
44
|
Constructor of the F5 class.
|
c9fdeb00
Chunk
LSB and F4 added.
|
45
|
"""
|
c9fdeb00
Chunk
LSB and F4 added.
|
46
|
StegBase.__init__(self, key)
|
8cfc1a23
Chunk
F5 half-finished.
|
47
|
self._embed_fun = None
|
c9fdeb00
Chunk
LSB and F4 added.
|
48
|
self.default_embedding = True
|
8cfc1a23
Chunk
F5 half-finished.
|
49
50
51
52
53
54
55
|
# needed because k is embedded separately
self.k_coeff = k
def _get_cov_data(self, img_path):
"""
|
c6c61f81
Chunk
staged.
|
56
|
Returns DCT coefficients of the cover image.
|
8cfc1a23
Chunk
F5 half-finished.
|
57
58
59
60
61
62
63
64
65
66
|
"""
self.cov_jpeg = jpegObj.Jpeg(img_path, key=self.key)
cov_data = self.cov_jpeg.getsignal(channel='Y')
self.cov_data = np.array(cov_data, dtype=np.int16)
return self.cov_data
def embed_raw_data(self, src_cover, src_hidden, tgt_stego, embed_fun='Default'):
"""This method embeds arbitrary data into a cover image.
The cover image must be a JPEG.
|
c9fdeb00
Chunk
LSB and F4 added.
|
67
68
69
70
|
@param embed_fun:
Specifies which embedding function should be used. Must be one of
'Default', 'F3', 'Jsteg'. If 'Default' is selected, the algorithm uses
|
8cfc1a23
Chunk
F5 half-finished.
|
71
|
the same behavior as Westfeld's implementation, i.e. decrementing
|
c9fdeb00
Chunk
LSB and F4 added.
|
72
73
|
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.
|
8cfc1a23
Chunk
F5 half-finished.
|
74
|
"""
|
c9fdeb00
Chunk
LSB and F4 added.
|
75
76
77
|
self.t0 = time.time()
if embed_fun == 'F3':
|
8cfc1a23
Chunk
F5 half-finished.
|
78
|
self._embed_fun = self._f3_embed
|
c9fdeb00
Chunk
LSB and F4 added.
|
79
80
|
self.default_embedding = False
elif embed_fun == 'JSteg' or embed_fun == 'LSB':
|
8cfc1a23
Chunk
F5 half-finished.
|
81
|
self._embed_fun = self._jsteg_embed
|
c9fdeb00
Chunk
LSB and F4 added.
|
82
83
84
|
self.default_embedding = False
else:
self._embed_fun = self._f3_embed
|
8cfc1a23
Chunk
F5 half-finished.
|
85
86
87
88
|
self.default_embedding = True
try:
cov_data = self._get_cov_data(src_cover)
|
c9fdeb00
Chunk
LSB and F4 added.
|
89
|
hid_data = self._get_hid_data(src_hidden)
|
8cfc1a23
Chunk
F5 half-finished.
|
90
|
# print hid_data.dtype,type(hid_data),hid_data.tolist()
|
c9fdeb00
Chunk
LSB and F4 added.
|
91
|
|
8cfc1a23
Chunk
F5 half-finished.
|
92
93
94
|
cov_data, bits_cnt = self._raw_embed(cov_data, hid_data)
if bits_cnt < np.size(hid_data) * 8:
|
c9fdeb00
Chunk
LSB and F4 added.
|
95
|
raise Exception("Expected embedded size is %db but actually %db." % (
|
8cfc1a23
Chunk
F5 half-finished.
|
96
97
|
np.size(hid_data) * 8, bits_cnt))
|
c9fdeb00
Chunk
LSB and F4 added.
|
98
|
self.cov_jpeg.setsignal(cov_data, channel='Y')
|
873557f9
Chunk
staged.
|
99
100
|
self.cov_jpeg.Jwrite(tgt_stego)
|
c9fdeb00
Chunk
LSB and F4 added.
|
101
|
# size_cov = os.path.getsize(tgt_stego)
|
873557f9
Chunk
staged.
|
102
103
104
105
106
107
108
|
size_cov = np.size(cov_data) / 8
size_embedded = np.size(hid_data)
self._display_stats("embedded", size_cov, size_embedded,
time.time() - self.t0)
except TypeError as e:
|
c9fdeb00
Chunk
LSB and F4 added.
|
109
|
raise e
|
8cfc1a23
Chunk
F5 half-finished.
|
110
111
112
113
114
115
116
117
|
except Exception as expt:
print "Exception when embedding!"
raise
def extract_raw_data(self, src_steg, tgt_hidden, embed_fun='Default'):
self.t0 = time.time()
|
c9fdeb00
Chunk
LSB and F4 added.
|
118
|
if embed_fun == 'F3':
|
8cfc1a23
Chunk
F5 half-finished.
|
119
120
121
122
123
124
|
self._embed_fun = self._f3_embed
self.default_embedding = False
elif embed_fun == 'JSteg' or embed_fun == 'LSB':
self._embed_fun = self._jsteg_embed
self.default_embedding = False
else:
|
c9fdeb00
Chunk
LSB and F4 added.
|
125
|
self._embed_fun = self._f3_embed
|
8cfc1a23
Chunk
F5 half-finished.
|
126
127
|
self.default_embedding = True
|
c9fdeb00
Chunk
LSB and F4 added.
|
128
129
|
try:
steg_data = self._get_cov_data(src_steg)
|
8cfc1a23
Chunk
F5 half-finished.
|
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
|
# emb_size = os.path.getsize(src_steg)
emb_size = np.size(steg_data) / 8
# recovering file size
header_size = 4 * 8
size_data, bits_cnt = self._raw_extract(steg_data, header_size)
if bits_cnt < header_size:
raise Exception("Expected embedded size is %db but actually %db." % (
header_size, bits_cnt))
size_data = bits2bytes(size_data[:header_size])
print size_data
size_hd = 0
for i in xrange(4):
size_hd += size_data[i] * 256 ** i
raw_size = size_hd * 8
if raw_size > np.size(steg_data):
raise Exception("Supposed secret data too large for stego image.")
hid_data, bits_cnt = self._raw_extract(steg_data, raw_size)
if bits_cnt < raw_size:
raise Exception("Expected embedded size is %db but actually %db." % (
raw_size, bits_cnt))
hid_data = bits2bytes(hid_data)
# print hid_data.dtype,type(hid_data),hid_data.tolist()
hid_data[4:].tofile(tgt_hidden)
self._display_stats("extracted", emb_size,
np.size(hid_data),
time.time() - self.t0)
except Exception as expt:
print "Exception when extracting!"
raise
|
c9fdeb00
Chunk
LSB and F4 added.
|
171
|
def _embed_k(self, cov_data, hid_data):
|
c9fdeb00
Chunk
LSB and F4 added.
|
172
173
174
175
176
177
178
179
180
181
182
183
184
|
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
|
033d3b0d
Chunk
staged.
|
185
|
if m != cov_data[self.dct_p[self.cov_ind]] & 1:
|
c9fdeb00
Chunk
LSB and F4 added.
|
186
187
188
189
190
191
192
193
194
195
196
197
198
|
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:
|
033d3b0d
Chunk
staged.
|
199
200
|
self.steg_ind += 1
k_split[i] = self.steg_data[self.dct_p[self.steg_ind]] & 1
|
c9fdeb00
Chunk
LSB and F4 added.
|
201
202
203
204
205
206
207
208
209
210
211
212
|
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
|
033d3b0d
Chunk
staged.
|
213
|
update_cnt -= 1
|
c9fdeb00
Chunk
LSB and F4 added.
|
214
215
216
217
218
219
220
221
222
223
224
225
226
|
# 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
|
033d3b0d
Chunk
staged.
|
227
|
while True:
|
c9fdeb00
Chunk
LSB and F4 added.
|
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
|
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] & 0xfffffffe) | m
|
8cfc1a23
Chunk
F5 half-finished.
|
243
|
def _raw_embed(self, cov_data, hid_data):
|
c9fdeb00
Chunk
LSB and F4 added.
|
244
|
k = self.k_coeff
|
8cfc1a23
Chunk
F5 half-finished.
|
245
|
n = (1 << k) - 1
|
c9fdeb00
Chunk
LSB and F4 added.
|
246
247
|
if n == 1 and self.default_embedding:
|
8cfc1a23
Chunk
F5 half-finished.
|
248
|
# in case k = n = 1, Westfeld's implementation uses F4 for embedding.
|
c9fdeb00
Chunk
LSB and F4 added.
|
249
|
f4 = F4(key=self.key)
|
8cfc1a23
Chunk
F5 half-finished.
|
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
|
return f4._raw_embed(cov_data, hid_data)
hid_data = bytes2bits(hid_data)
if len(hid_data) % k != 0:
hid_data = list(hid_data) + [0 for x in range(k - len(hid_data) % k)]
ind_nonzero = np.nonzero(cov_data)[0]
if np.size(ind_nonzero) * k < len(hid_data) * n:
raise Exception("Supposed secret data too large for stego image.")
ind_cov = 0
for ind_hid in range(0, len(hid_data), k):
msg_chunk = hid_data[ind_hid:ind_hid + k]
cov_chunk = ind_nonzero[ind_cov:ind_cov + n]
ind_cov += n
success = False
|
c9fdeb00
Chunk
LSB and F4 added.
|
268
269
|
while not success:
h = 0
|
8cfc1a23
Chunk
F5 half-finished.
|
270
|
for i in xrange(n):
|
c9fdeb00
Chunk
LSB and F4 added.
|
271
272
273
274
275
|
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] # N.B. hid_data[0]:high (that is x2), hid_data[1]:low (that is x1)
|
8cfc1a23
Chunk
F5 half-finished.
|
276
277
|
s = scalar_x ^ h
if s != 0:
|
c9fdeb00
Chunk
LSB and F4 added.
|
278
279
280
281
282
283
|
self._embed_fun(cov_data, cov_chunk[s - 1])
else:
break
if cov_data[cov_chunk[s - 1]] == 0: # shrinkage
cov_chunk[s - 1:-1] = cov_chunk[s:]
|
8cfc1a23
Chunk
F5 half-finished.
|
284
285
286
287
|
cov_chunk[-1] = ind_nonzero[ind_cov]
ind_cov += 1
else:
success = True
|
c9fdeb00
Chunk
LSB and F4 added.
|
288
289
290
|
return cov_data, ind_hid + k
|
8cfc1a23
Chunk
F5 half-finished.
|
291
|
def _raw_extract(self, steg_data, num_bits):
|
c9fdeb00
Chunk
LSB and F4 added.
|
292
|
k = self.k_coeff
|
8cfc1a23
Chunk
F5 half-finished.
|
293
|
n = (1 << k) - 1
|
c9fdeb00
Chunk
LSB and F4 added.
|
294
295
|
if n == 1 and self.default_embedding:
|
c9fdeb00
Chunk
LSB and F4 added.
|
296
|
f4 = F4(key=self.key)
|
8cfc1a23
Chunk
F5 half-finished.
|
297
298
299
|
return f4._raw_extract(steg_data, num_bits)
num_bits_ceil = num_bits
|
c9fdeb00
Chunk
LSB and F4 added.
|
300
|
if num_bits % k != 0:
|
8cfc1a23
Chunk
F5 half-finished.
|
301
302
303
|
num_bits_ceil = k * (num_bits / k + 1)
hid_data = np.zeros(num_bits_ceil, np.uint8)
|
c9fdeb00
Chunk
LSB and F4 added.
|
304
|
curr_chunk = np.zeros(k, np.uint8)
|
8cfc1a23
Chunk
F5 half-finished.
|
305
306
307
308
309
310
|
steg_data = steg_data[np.nonzero(steg_data)]
ind_hid = 0
for ind_cov in range(0, len(steg_data), n):
steg_chunk = steg_data[ind_cov:ind_cov + n]
h = 0 # hash value
|
c9fdeb00
Chunk
LSB and F4 added.
|
311
312
313
314
315
316
|
for i in xrange(n):
h ^= ((steg_chunk[i] & 1) * (i + 1))
for i in xrange(k):
curr_chunk[k - i - 1] = h & 1 # N.B. hid_data[0]:high (that is x2), hid_data[1]:low (that is x1)
h >>= 1
|
8cfc1a23
Chunk
F5 half-finished.
|
317
318
|
hid_data[ind_hid:ind_hid + k] = curr_chunk[0:k]
|
c9fdeb00
Chunk
LSB and F4 added.
|
319
|
ind_hid += k
|
8cfc1a23
Chunk
F5 half-finished.
|
320
321
|
if ind_hid >= num_bits_ceil: break
|
c9fdeb00
Chunk
LSB and F4 added.
|
322
|
|
8cfc1a23
Chunk
F5 half-finished.
|
323
|
return hid_data, num_bits_ceil
|
c9fdeb00
Chunk
LSB and F4 added.
|
324
|
|
8cfc1a23
Chunk
F5 half-finished.
|
325
|
def __str__(self):
|
c9fdeb00
Chunk
LSB and F4 added.
|
326
|
return 'F5'
|