Commit c9fdeb00fb59fd89b2d92c58636b27ad186ea036
1 parent
548d95dc
Exists in
master
LSB and F4 added.
Showing
5 changed files
with
545 additions
and
193 deletions
Show diff stats
msteg/steganography/F3.py
| ... | ... | @@ -0,0 +1,99 @@ |
| 1 | +""" | |
| 2 | +<p>This module implements a slight variant of the F4 steganography algorithm | |
| 3 | +invented by Andreas Westfeld. It embeds a secret message in JPEG | |
| 4 | +DCT coefficients.</p> | |
| 5 | +It differs from F3 in that even negative and odd positive DCT | |
| 6 | +coefficients represent a 1 and odd negative and even positive | |
| 7 | +DCT coefficients represent a 0. It also supports permutative strattling | |
| 8 | +which is not included in the original description of F4. | |
| 9 | +""" | |
| 10 | +import time | |
| 11 | +import numpy as np | |
| 12 | +from StegBase import * | |
| 13 | +from common import * | |
| 14 | + | |
| 15 | + | |
| 16 | +class F3(StegBase): | |
| 17 | + """ This module has two methods: <i>embed_raw_data</i> to embed data | |
| 18 | + with the F3 algorithm and <i>extract_raw_data</i> to extract data | |
| 19 | + which was embedded previously. """ | |
| 20 | + | |
| 21 | + def __init__(self): | |
| 22 | + """ | |
| 23 | + Constructor of the F3 class. | |
| 24 | + """ | |
| 25 | + StegBase.__init__(self) | |
| 26 | + | |
| 27 | + def embed_raw_data(self, src_cover, src_hidden, tgt_stego): | |
| 28 | + """ This method embeds arbitrary data into a cover image. | |
| 29 | + The cover image must be a JPEG. | |
| 30 | + | |
| 31 | + src_cover - A valid pathname to an image file which serves as cover image | |
| 32 | + (the image which the secret image is embedded into). | |
| 33 | + | |
| 34 | + src_hidden - A valid pathname to an arbitrary file that is supposed to be | |
| 35 | + embedded into the cover image. | |
| 36 | + | |
| 37 | + tgt_stego - Target pathname of the resulting stego image. You should save to a | |
| 38 | + PNG or another lossless format, because many LSBs don't survive | |
| 39 | + lossy compression. | |
| 40 | + """ | |
| 41 | + self.t0 = time.time() | |
| 42 | + StegBase._post_embed_actions(self, src_cover, src_hidden, tgt_stego) | |
| 43 | + | |
| 44 | + def extract_raw_data(self, src_steg, tgt_hidden): | |
| 45 | + """ This method extracts secret data from a stego image. It is | |
| 46 | + (obviously) the inverse operation of embed_raw_data. | |
| 47 | + | |
| 48 | + src_stego - A valid pathname to an image file which serves as stego image. | |
| 49 | + | |
| 50 | + tgt_hidden - A pathname denoting where the extracted data should be saved to. | |
| 51 | + """ | |
| 52 | + self.t0 = time.time() | |
| 53 | + StegBase._post_extract_actions(self, src_steg, tgt_hidden) | |
| 54 | + | |
| 55 | + def _raw_embed(self, cov_data, hid_data, status_begin=0): | |
| 56 | + """ | |
| 57 | + cov_data - 4-D numpy.int32 array | |
| 58 | + hid_data - 1-D numpy.uint8 array | |
| 59 | + """ | |
| 60 | + hid_data = bytes2bits(hid_data) | |
| 61 | + i = 0 | |
| 62 | + cnt = -1 | |
| 63 | + for x in np.nditer(cov_data, op_flags=['readwrite']): | |
| 64 | + cnt = cnt + 1 | |
| 65 | + if x == 0 or cnt % 64 == 0: continue | |
| 66 | + | |
| 67 | + m = (hid_data[i] & 1) | |
| 68 | + if x > 0 and x & 1 != m: | |
| 69 | + x[...] -= 1 | |
| 70 | + elif x < 0 and x & 1 == m: | |
| 71 | + x[...] += 1 | |
| 72 | + if x == 0: continue | |
| 73 | + i += 1 | |
| 74 | + if i == hid_data.size: break | |
| 75 | + | |
| 76 | + return cov_data | |
| 77 | + | |
| 78 | + def _raw_extract(self, steg_data, num_bits): | |
| 79 | + """ | |
| 80 | + Just a small helper function to extract hidden data. | |
| 81 | + """ | |
| 82 | + hid_data = np.zeros(num_bits, np.uint8) | |
| 83 | + j = 0 | |
| 84 | + cnt = -1 | |
| 85 | + for x in np.nditer(steg_data): | |
| 86 | + cnt = cnt + 1 | |
| 87 | + if x == 0 or cnt % 64 == 0: continue | |
| 88 | + if j >= num_bits: break | |
| 89 | + if x > 0: | |
| 90 | + hid_data[j] = x & 1 | |
| 91 | + else: | |
| 92 | + hid_data[j] = (x & 1) ^ 1 | |
| 93 | + | |
| 94 | + j = j + 1 | |
| 95 | + | |
| 96 | + return hid_data | |
| 97 | + | |
| 98 | + def __str__(self): | |
| 99 | + return "F4'" | ... | ... |
| ... | ... | @@ -0,0 +1,372 @@ |
| 1 | +""" | |
| 2 | +<p>This module implements the rather sophisticated F5 algorithm which was | |
| 3 | +invented by Andreas Westfeld.</p> | |
| 4 | + | |
| 5 | +Unlike its vastly inferior predecessors, namely F3 and F4, it features matrix | |
| 6 | +encoding which makes it possible to embed a chunk of k bits within 2^k - 1 | |
| 7 | +bits of the cover data and only change one bit (at most). A bit change is | |
| 8 | +done by subtracting the absolute value of the corresponding DCT coefficient. | |
| 9 | +When the embedding process begins, the parameter k is computed based on | |
| 10 | +the capacity of the cover image and the prospective embedding ratio. | |
| 11 | +With small amount of hidden data k becomes large which leads to a greater | |
| 12 | +embedding efficiency (embedded information per bit change).<br /> | |
| 13 | + | |
| 14 | +A permutation (initialized by a user-supplied seed) of the DCT coefficients | |
| 15 | +helps to scatter each chunk across the entire image. | |
| 16 | +F5 can be seen as meta-algorithm as it uses a coding scheme to change | |
| 17 | +as little data as possible and then applies a simpler algorithm (such as F3) | |
| 18 | +to actually embed data. That is why this module allows the user to specify | |
| 19 | +which embedding function (one of JSteg, F3, F4) should be used. | |
| 20 | +""" | |
| 21 | + | |
| 22 | +import time | |
| 23 | +import math | |
| 24 | +import numpy as np | |
| 25 | +from stegotool.plugins.steganography.F4.F4 import F4 | |
| 26 | +from stegotool.util.JPEGSteg import JPEGSteg | |
| 27 | +from stegotool.util.plugins import describe_annotate_convert | |
| 28 | +from stegotool.util.plugins import ident, ImagePath, FilePath, NewFilePath | |
| 29 | + | |
| 30 | + | |
| 31 | +class F5(JPEGSteg): | |
| 32 | + """ This module has two methods: <i>embed_raw_data</i> to embed data | |
| 33 | + with the F5 algorithm and <i>extract_raw_data</i> to extract data | |
| 34 | + which was embedded previously. """ | |
| 35 | + | |
| 36 | + def __init__(self, ui, core): | |
| 37 | + """ | |
| 38 | + Constructor of the F5 class. | |
| 39 | + """ | |
| 40 | + JPEGSteg.__init__(self, ui, core) | |
| 41 | + self._embed_hook = self._embed_k | |
| 42 | + self._extract_hook = self._extract_k | |
| 43 | + self._embed_fun = None | |
| 44 | + self.dct_p = None | |
| 45 | + self.seed = None | |
| 46 | + self.default_embedding = True | |
| 47 | + self.steg_ind = -1 | |
| 48 | + self.excess_bits = None | |
| 49 | + # needed because k is embedded separately | |
| 50 | + self.cov_ind = -1 | |
| 51 | + self.k_coeff = -1 | |
| 52 | + | |
| 53 | + @describe_annotate_convert((None, None, ident), | |
| 54 | + ("cover image", ImagePath, str), | |
| 55 | + ("hidden data", FilePath, str), | |
| 56 | + ("stego image", NewFilePath, str), | |
| 57 | + ("seed", int, int), | |
| 58 | + ("embedding behavior", | |
| 59 | + ['Default', 'F3', 'JSteg'], str)) | |
| 60 | + def embed_raw_data(self, src_cover, src_hidden, tgt_stego, seed, | |
| 61 | + embed_fun): | |
| 62 | + """<p>This method embeds arbitrary data into a cover image. | |
| 63 | + The cover image must be a JPEG.</p> | |
| 64 | + | |
| 65 | + <p>Parameters: | |
| 66 | + <ol> | |
| 67 | + <li><pre>src_cover</pre> | |
| 68 | + A valid pathname to an image file which serves as cover image | |
| 69 | + (the image which the secret image is embedded into).</li> | |
| 70 | + | |
| 71 | + <li><pre>src_hidden</pre> | |
| 72 | + A valid pathname to an arbitrary file that is supposed to be | |
| 73 | + embedded into the cover image.</li> | |
| 74 | + | |
| 75 | + <li><pre>tgt_stego</pre> | |
| 76 | + Target pathname of the resulting stego image. You should save to | |
| 77 | + a PNG or another lossless format, because many LSBs don't survive | |
| 78 | + lossy compression.</li> | |
| 79 | + | |
| 80 | + <li><pre>seed</pre> | |
| 81 | + A seed for the random number generator that is responsible scattering | |
| 82 | + the secret data within the cover image.</li> | |
| 83 | + | |
| 84 | + <li><pre>param embed_fun</pre> | |
| 85 | + Specifies which embedding function should be used. Must be one of | |
| 86 | + 'Default', 'F3', 'Jsteg'. If 'Default' is selected, the algorithm uses | |
| 87 | + the same behavior as Westfeld's implementation, i.e. decrementing | |
| 88 | + absolute values for n > 1 (F3) and using F4 in the special case n = 1. | |
| 89 | + Selecting F3 or JSteg results in using that scheme for all n.</li> | |
| 90 | + </ol> | |
| 91 | + </p> | |
| 92 | + """ | |
| 93 | + self.t0 = time.time() | |
| 94 | + self.seed = seed | |
| 95 | + if embed_fun == 'F3': | |
| 96 | + self._embed_fun = self._f3_embed | |
| 97 | + self.default_embedding = False | |
| 98 | + elif embed_fun == 'JSteg': | |
| 99 | + self._embed_fun = self._jsteg_embed | |
| 100 | + self.default_embedding = False | |
| 101 | + elif embed_fun == 'Default': | |
| 102 | + self._embed_fun = self._f3_embed | |
| 103 | + self.default_embedding = True | |
| 104 | + | |
| 105 | + self.cov_ind = -1 | |
| 106 | + JPEGSteg._post_embed_actions(self, src_cover, src_hidden, tgt_stego) | |
| 107 | + | |
| 108 | + @describe_annotate_convert((None, None, ident), | |
| 109 | + ("stego image", ImagePath, str), | |
| 110 | + ("hidden data", NewFilePath, str), | |
| 111 | + ("seed", int, int), | |
| 112 | + ("embedding behavior", ['Default', 'F3/JSteg'], | |
| 113 | + str)) | |
| 114 | + def extract_raw_data(self, src_steg, tgt_hidden, seed, embed_fun): | |
| 115 | + """<p>This method extracts secret data from a stego image. It is | |
| 116 | + (obviously) the inverse operation of embed_raw_data.</p> | |
| 117 | + | |
| 118 | + <p>Parameters: | |
| 119 | + <ol> | |
| 120 | + <li><pre>src_stego</pre> | |
| 121 | + A valid pathname to an image file which serves as stego image.</li> | |
| 122 | + | |
| 123 | + <li><pre>tgt_hidden</pre> | |
| 124 | + A pathname denoting where the extracted data should be saved to.</li> | |
| 125 | + | |
| 126 | + <li><pre>param seed</pre> | |
| 127 | + A seed for the random number generator that is responsible scattering | |
| 128 | + the secret data within the cover image.</li> | |
| 129 | + | |
| 130 | + <li><pre>param embed_fun</pre> | |
| 131 | + Specifies which embedding function should be used. Must be one of | |
| 132 | + 'Default', 'F3', 'JSteg'. If 'Default' is selected, the algorithm uses | |
| 133 | + the same behavior as Westfeld's implementation, i.e. decrementing | |
| 134 | + absolute values for n > 1 (F3) and using F4 in the special case n = 1. | |
| 135 | + Selecting F3 or JSteg results in using that scheme for all n.</li> | |
| 136 | + </ol></pre> | |
| 137 | + """ | |
| 138 | + | |
| 139 | + self.t0 = time.time() | |
| 140 | + self.seed = seed | |
| 141 | + self.steg_ind = -1 | |
| 142 | + if embed_fun == 'F3/JSteg': | |
| 143 | + self.default_embedding = False | |
| 144 | + elif embed_fun == 'Default': | |
| 145 | + self.default_embedding = True | |
| 146 | + | |
| 147 | + # excess bits occur when the size of extracted data is not a multiple | |
| 148 | + # of k. if excess bits are available, they are prepended to hidden data | |
| 149 | + self.excess_bits = None | |
| 150 | + | |
| 151 | + JPEGSteg._post_extract_actions(self, src_steg, tgt_hidden) | |
| 152 | + | |
| 153 | + def _embed_k(self, cov_data, hid_data): | |
| 154 | + np.random.seed(self.seed) | |
| 155 | + self.dct_p = np.random.permutation(cov_data.size) | |
| 156 | + self.k_coeff = self._find_max_k(cov_data, hid_data) | |
| 157 | + self.ui.display_status('setting k = %d' % self.k_coeff) | |
| 158 | + k_split = self.lookup_tab.split_byte(self.k_coeff, 1)[-4:] | |
| 159 | + # embed k in F3-like style | |
| 160 | + for m in k_split: | |
| 161 | + success = False | |
| 162 | + while not success: | |
| 163 | + self.cov_ind += 1 | |
| 164 | + while cov_data[self.dct_p[self.cov_ind]] == 0 or \ | |
| 165 | + self.dct_p[self.cov_ind] % 64 == 0: | |
| 166 | + self.cov_ind += 1 | |
| 167 | + if m != cov_data[self.dct_p[self.cov_ind]] & 1: | |
| 168 | + cov_data[self.dct_p[self.cov_ind]] -= \ | |
| 169 | + math.copysign(1, cov_data[self.dct_p[self.cov_ind]]) | |
| 170 | + success = cov_data[self.dct_p[self.cov_ind]] != 0 | |
| 171 | + | |
| 172 | + def _extract_k(self, steg_data): | |
| 173 | + # initializing the MT is done only once in order to retain the state | |
| 174 | + self.dct_p = np.random.seed(self.seed) | |
| 175 | + self.dct_p = np.random.permutation(self.steg_data.size) | |
| 176 | + k_split = np.zeros(4, np.uint8) | |
| 177 | + for i in xrange(k_split.size): | |
| 178 | + self.steg_ind += 1 | |
| 179 | + while self.steg_data[self.dct_p[self.steg_ind]] == 0 or\ | |
| 180 | + self.dct_p[self.steg_ind] % 64 == 0: | |
| 181 | + self.steg_ind += 1 | |
| 182 | + k_split[i] = self.steg_data[self.dct_p[self.steg_ind]] & 1 | |
| 183 | + self.k_coeff = self.lookup_tab.merge_words(tuple([0, 0, 0, 0] + | |
| 184 | + list(k_split)), 1) | |
| 185 | + | |
| 186 | + def _find_max_k(self, cov_data, hid_data): | |
| 187 | + cnt = 4 # information about k take up 4 bits | |
| 188 | + # find number of DCT coefficients | |
| 189 | + update_cnt = 10000 | |
| 190 | + for i, c in enumerate(cov_data): | |
| 191 | + if update_cnt == 0: | |
| 192 | + self._set_progress( | |
| 193 | + int(30 * (float(i) / float(cov_data.size)))) | |
| 194 | + update_cnt = 10000 | |
| 195 | + update_cnt -= 1 | |
| 196 | + # pessimistic, but accurate estimation of the capacity of the image | |
| 197 | + ci = int(c) | |
| 198 | + if (not (ci is 0)) and (not ((i % 64) is 0)) \ | |
| 199 | + and (not (ci is 1)) and (not (ci is -1)): | |
| 200 | + cnt += 1 | |
| 201 | + hid_size = hid_data.size | |
| 202 | + cov_size = cnt | |
| 203 | + if cov_size < hid_size: | |
| 204 | + raise Exception("Cannot fit %d bits in %d DCT coefficients. \ | |
| 205 | + Cover image is too small." % (hid_size, cov_size)) | |
| 206 | + self.ui.display_status('DCT embedding ratio = %f' \ | |
| 207 | + % (float(hid_size) / float(cov_size))) | |
| 208 | + k = 1 | |
| 209 | + while True: | |
| 210 | + k += 1 | |
| 211 | + n = (1 << k) - 1 | |
| 212 | + num_chunks = cov_size / n | |
| 213 | + num_emb_bits = num_chunks * k | |
| 214 | + if num_emb_bits < hid_size: | |
| 215 | + return min(k - 1, 15) | |
| 216 | + | |
| 217 | + # low level embedding functions | |
| 218 | + def _f3_embed(self, cov_data, ind): | |
| 219 | + cov_data[ind] -= math.copysign(1, cov_data[ind]) | |
| 220 | + | |
| 221 | + def _jsteg_embed(self, cov_data, ind): | |
| 222 | + m = 1 ^ (cov_data[ind] & 1) | |
| 223 | + cov_data[ind] = (cov_data[ind] & 0xffffe) | m | |
| 224 | + | |
| 225 | + def _raw_embed(self, cov_data, hid_data, status_begin=0): | |
| 226 | + k = self.k_coeff | |
| 227 | + n = (1 << k) - 1 | |
| 228 | + if n == 1 and self.default_embedding: | |
| 229 | + # in case k = n = 1, Westfeld's implementation uses F4 for | |
| 230 | + # embedding. Therefore, if 'default' embedding has been selected | |
| 231 | + # we will do the same | |
| 232 | + f4 = F4(self.ui, self.core) | |
| 233 | + f4.seed = self.seed | |
| 234 | + f4.dct_p = self.dct_p | |
| 235 | + f4.cov_ind = self.cov_ind | |
| 236 | + cov_data = f4._raw_embed(cov_data, hid_data, 30) | |
| 237 | + return cov_data | |
| 238 | + | |
| 239 | + cov_ind = self.cov_ind # preventing RSI by writing 'self' less often | |
| 240 | + hid_ind = 0 | |
| 241 | + remaining_bits = hid_data.size | |
| 242 | + hid_size = float(hid_data.size) | |
| 243 | + dct_p = self.dct_p | |
| 244 | + | |
| 245 | + update_cnt = int(hid_size / (70.0 * k)) | |
| 246 | + while remaining_bits > 0: | |
| 247 | + if update_cnt == 0: | |
| 248 | + self._set_progress(30 + int((( | |
| 249 | + hid_size - remaining_bits) / hid_size) * 70)) | |
| 250 | + update_cnt = int(hid_size / (70.0 * k)) | |
| 251 | + update_cnt -= 1 | |
| 252 | + msg_chunk_size = min(remaining_bits, k) | |
| 253 | + msg_chunk = np.zeros(k, np.int8) | |
| 254 | + cov_chunk = np.zeros(n, np.int32) | |
| 255 | + msg_chunk[0:msg_chunk_size] = hid_data[hid_ind:hid_ind + | |
| 256 | + msg_chunk_size] | |
| 257 | + hid_ind += k | |
| 258 | + | |
| 259 | + # get n DCT coefficients | |
| 260 | + for i in xrange(n): | |
| 261 | + cov_ind += 1 | |
| 262 | + while cov_data[dct_p[cov_ind]] == 0 \ | |
| 263 | + or dct_p[cov_ind] % 64 == 0: | |
| 264 | + cov_ind += 1 | |
| 265 | + cov_chunk[i] = dct_p[cov_ind] | |
| 266 | + | |
| 267 | + success = False | |
| 268 | + while not success: # loop necessary because of shrinkage | |
| 269 | + h = 0 | |
| 270 | + for i in xrange(n): | |
| 271 | + h ^= ((cov_data[cov_chunk[i]] & 1) * (i + 1)) | |
| 272 | + scalar_x = 0 | |
| 273 | + for i in xrange(k): | |
| 274 | + scalar_x = (scalar_x << 1) + msg_chunk[i] | |
| 275 | + s = scalar_x ^ h | |
| 276 | + if s != 0: | |
| 277 | + self._embed_fun(cov_data, cov_chunk[s - 1]) | |
| 278 | + else: | |
| 279 | + break | |
| 280 | + | |
| 281 | + if cov_data[cov_chunk[s - 1]] == 0: # test for shrinkage | |
| 282 | + cov_chunk[s - 1:-1] = cov_chunk[s:] # adjusting | |
| 283 | + cov_ind += 1 | |
| 284 | + while cov_data[dct_p[cov_ind]] == 0 or\ | |
| 285 | + dct_p[cov_ind] % 64 == 0: | |
| 286 | + cov_ind += 1 | |
| 287 | + cov_chunk[n - 1] = dct_p[cov_ind] | |
| 288 | + else: | |
| 289 | + success = True | |
| 290 | + | |
| 291 | + remaining_bits -= k | |
| 292 | + | |
| 293 | + self.k_coeff = -1 # prevent k being read from this instance | |
| 294 | + return cov_data | |
| 295 | + | |
| 296 | + def _raw_extract(self, num_bits): | |
| 297 | + k = self.k_coeff | |
| 298 | + n = (1 << k) - 1 | |
| 299 | + if self.is_header == None: | |
| 300 | + self.is_header = True | |
| 301 | + if n == 1 and self.default_embedding: | |
| 302 | + f4 = F4(self.ui, self.core) | |
| 303 | + f4.seed = self.seed | |
| 304 | + f4.dct_p = self.dct_p | |
| 305 | + f4.steg_data = self.steg_data | |
| 306 | + f4.is_header = self.is_header | |
| 307 | + f4.steg_ind = self.steg_ind | |
| 308 | + hid_data = f4._raw_extract(num_bits) | |
| 309 | + self.steg_ind = f4.steg_ind | |
| 310 | + self.is_header = False | |
| 311 | + return hid_data | |
| 312 | + remaining_bits = num_bits | |
| 313 | + hid_data = np.zeros(num_bits, np.uint8) | |
| 314 | + hid_ind = 0 | |
| 315 | + | |
| 316 | + dct_p = self.dct_p | |
| 317 | + | |
| 318 | + is_header = False # signals whether or not extracting header | |
| 319 | + | |
| 320 | + if self.excess_bits != None: | |
| 321 | + hid_data[hid_ind:hid_ind + self.excess_bits.size] = \ | |
| 322 | + self.excess_bits | |
| 323 | + hid_ind += self.excess_bits.size | |
| 324 | + remaining_bits -= self.excess_bits.size | |
| 325 | + | |
| 326 | + curr_chunk = np.zeros(k, np.uint8) | |
| 327 | + | |
| 328 | + update_cnt = int(num_bits / (100.0 * k)) | |
| 329 | + | |
| 330 | + while remaining_bits > 0: | |
| 331 | + | |
| 332 | + if update_cnt == 0 and not is_header: | |
| 333 | + self._set_progress(int(((float(num_bits) \ | |
| 334 | + - remaining_bits) / num_bits) * 100)) | |
| 335 | + update_cnt = int(num_bits / (100.0 * k)) | |
| 336 | + | |
| 337 | + update_cnt -= 1 | |
| 338 | + | |
| 339 | + steg_chunk = [0 for i in xrange(n)] | |
| 340 | + for i in xrange(n): | |
| 341 | + self.steg_ind += 1 | |
| 342 | + while self.steg_data[dct_p[self.steg_ind]] == 0 or\ | |
| 343 | + dct_p[self.steg_ind] % 64 == 0: | |
| 344 | + self.steg_ind += 1 | |
| 345 | + steg_chunk[i] = self.steg_data[dct_p[self.steg_ind]] | |
| 346 | + | |
| 347 | + h = 0 # hash value | |
| 348 | + for i in xrange(n): | |
| 349 | + h ^= ((steg_chunk[i] & 1) * (i + 1)) | |
| 350 | + | |
| 351 | + for i in xrange(k): | |
| 352 | + curr_chunk[k - i - 1] = h % 2 | |
| 353 | + h /= 2 | |
| 354 | + | |
| 355 | + l = min(k, remaining_bits) | |
| 356 | + for i in xrange(l): | |
| 357 | + hid_data[hid_ind] = curr_chunk[i] | |
| 358 | + hid_ind += 1 | |
| 359 | + | |
| 360 | + # save excess bits (for later calls) | |
| 361 | + if k > remaining_bits: | |
| 362 | + self.excess_bits = curr_chunk[remaining_bits:] | |
| 363 | + else: | |
| 364 | + self.excess_bits = None | |
| 365 | + | |
| 366 | + remaining_bits -= k | |
| 367 | + | |
| 368 | + self.is_header = False | |
| 369 | + return hid_data | |
| 370 | + | |
| 371 | + def __str__(self): | |
| 372 | + return 'F5' | ... | ... |
msteg/steganography/LSB.py
| ... | ... | @@ -10,203 +10,83 @@ bits a parameter named word_size. Thus --- in this context --- word means |
| 10 | 10 | import time |
| 11 | 11 | import numpy as np |
| 12 | 12 | import scipy as sp |
| 13 | -from stegotool.util.lookup import Lookup | |
| 14 | -from stegotool.util.plugins import StegoBase | |
| 15 | -from stegotool.util.plugins import describe_annotate_convert | |
| 16 | -from stegotool.util.plugins import ident, ImagePath, FilePath, NewFilePath | |
| 17 | -import Image | |
| 18 | - | |
| 19 | - | |
| 20 | -class LSB(StegoBase): | |
| 21 | - """ This module has 4 methods: | |
| 22 | - <ul> | |
| 23 | - <li><i>embed_image</i> and <i>extract_image</i> to embed/extract | |
| 24 | - images.</li> | |
| 25 | - <li><i>embed_raw_data</i> and <i>extract_raw_data</i> to embed/extract | |
| 26 | - any data.</li> | |
| 27 | - </ul> | |
| 28 | - """ | |
| 13 | +from StegBase import * | |
| 14 | +from common import * | |
| 15 | + | |
| 16 | + | |
| 17 | +class LSB(StegBase): | |
| 18 | + """ This module has two methods: <i>embed_raw_data</i> to embed data | |
| 19 | + with the F3 algorithm and <i>extract_raw_data</i> to extract data | |
| 20 | + which was embedded previously. """ | |
| 21 | + | |
| 22 | + def __init__(self): | |
| 23 | + """ | |
| 24 | + Constructor of the F3 class. | |
| 25 | + """ | |
| 26 | + StegBase.__init__(self) | |
| 27 | + | |
| 28 | + def embed_raw_data(self, src_cover, src_hidden, tgt_stego): | |
| 29 | + """ This method embeds arbitrary data into a cover image. | |
| 30 | + The cover image must be a JPEG. | |
| 31 | + | |
| 32 | + src_cover - A valid pathname to an image file which serves as cover image | |
| 33 | + (the image which the secret image is embedded into). | |
| 34 | + | |
| 35 | + src_hidden - A valid pathname to an arbitrary file that is supposed to be | |
| 36 | + embedded into the cover image. | |
| 29 | 37 | |
| 30 | - def __init__(self, ui, core): | |
| 38 | + tgt_stego - Target pathname of the resulting stego image. You should save to a | |
| 39 | + PNG or another lossless format, because many LSBs don't survive | |
| 40 | + lossy compression. | |
| 31 | 41 | """ |
| 32 | - Constructor of the LSB class. | |
| 42 | + self.t0 = time.time() | |
| 43 | + StegBase._post_embed_actions(self, src_cover, src_hidden, tgt_stego) | |
| 44 | + | |
| 45 | + def extract_raw_data(self, src_steg, tgt_hidden): | |
| 46 | + """ This method extracts secret data from a stego image. It is | |
| 47 | + (obviously) the inverse operation of embed_raw_data. | |
| 48 | + | |
| 49 | + src_stego - A valid pathname to an image file which serves as stego image. | |
| 50 | + | |
| 51 | + tgt_hidden - A pathname denoting where the extracted data should be saved to. | |
| 52 | + """ | |
| 53 | + self.t0 = time.time() | |
| 54 | + StegBase._post_extract_actions(self, src_steg, tgt_hidden) | |
| 55 | + | |
| 56 | + def _raw_embed(self, cov_data, hid_data, status_begin=0): | |
| 33 | 57 | """ |
| 34 | - super(LSB, self).__init__(ui, core) | |
| 35 | - self.lookup_tab = Lookup(True) | |
| 36 | - | |
| 37 | - @describe_annotate_convert((None, None, ident), | |
| 38 | - ("cover image", ImagePath, str), | |
| 39 | - ("hidden data", FilePath, str), | |
| 40 | - ("stego image", NewFilePath, str), | |
| 41 | - ("word size", [1, 2, 4, 8], int)) | |
| 42 | - def embed_raw_data(self, src_cover, src_hidden, tgt_stego, word_size): | |
| 43 | - """ <p>This method embeds arbitrary data into a cover image. | |
| 44 | - Note that the cover image is of course decoded | |
| 45 | - before embedding, the secret data however is not.</p> | |
| 46 | - | |
| 47 | - <p>Parameters: | |
| 48 | - <ol> | |
| 49 | - <li><pre>src_cover</pre> | |
| 50 | - A valid pathname to an image file which serves as cover image | |
| 51 | - (the image which the secret image is embedded into).</li> | |
| 52 | - | |
| 53 | - <li><pre>src_hidden</pre> | |
| 54 | - A valid pathname to an arbitrary file that is supposed to be | |
| 55 | - embedded into the cover image.</li> | |
| 56 | - | |
| 57 | - <li><pre>tgt_stego</pre> | |
| 58 | - Target pathname of the resulting stego image. You should save to | |
| 59 | - a PNG or another lossless format, because many LSBs don't survive | |
| 60 | - lossy compression.</li> | |
| 61 | - | |
| 62 | - <li><pre>word_size</pre> | |
| 63 | - Must be an even divisor of 8, i.e. one of 1, 2, 4, 8. Specifies how | |
| 64 | - many least significant bits of each byte in the cover image are used | |
| 65 | - for embedding the secret data. The larger this number the more easily | |
| 66 | - successful steganalysis can be carried out.</li> | |
| 67 | - | |
| 68 | - <li><pre>scatter</pre> | |
| 69 | - If true, bytes used for embedding will be evenly distributed over the | |
| 70 | - entire cover image. This(slightly) complicates steganalysis. If false, | |
| 71 | - consecutive pixels are used for embedding, beginning with the first | |
| 72 | - byte in the cover image, until the secret data is exhausted.</li> | |
| 73 | - </ol> | |
| 74 | - An exception is raised if the secret data is too | |
| 75 | - large to be embedded into the cover image.</p> | |
| 58 | + cov_data - 4-D numpy.int32 array | |
| 59 | + hid_data - 1-D numpy.uint8 array | |
| 60 | + """ | |
| 61 | + hid_data = bytes2bits(hid_data) | |
| 62 | + i = 0 | |
| 63 | + cnt = -1 | |
| 64 | + for x in np.nditer(cov_data, op_flags=['readwrite']): | |
| 65 | + cnt = cnt + 1 | |
| 66 | + if x == 0 or x == 1 or cnt % 64 == 0: continue | |
| 67 | + | |
| 68 | + m = (hid_data[i] & 1) | |
| 69 | + x[...] = (x & 0xfffffffe) | m | |
| 70 | + i += 1 | |
| 71 | + if i == hid_data.size: break | |
| 72 | + | |
| 73 | + return cov_data | |
| 74 | + | |
| 75 | + def _raw_extract(self, steg_data, num_bits): | |
| 76 | 76 | """ |
| 77 | - t0 = time.time() | |
| 78 | - self.ui.set_progress(1) | |
| 79 | - word_size = int(word_size) | |
| 80 | - cov_img = self.core.media_manager.get_file(src_cover) | |
| 81 | - if cov_img and isinstance(cov_img, Image.Image): | |
| 82 | - cov_data = sp.misc.fromimage(cov_img) | |
| 83 | - elif cov_img and hasattr(cov_img, "data"): | |
| 84 | - raise Exception("cover image must be an image") | |
| 85 | - else: | |
| 86 | - cov_data = sp.misc.fromimage(Image.open(src_cover)) | |
| 87 | - | |
| 88 | - orig_shape = cov_data.shape | |
| 89 | - cov_data = cov_data.ravel() | |
| 90 | - | |
| 91 | - hidden_data_suffix = self.core.media_manager.get_file(src_hidden) | |
| 92 | - if hidden_data_suffix and hasattr(hidden_data_suffix, 'data'): | |
| 93 | - hidden_data_suffix = np.fromstring(hidden_data_suffix.data, | |
| 94 | - np.uint8) | |
| 95 | - elif hidden_data_suffix and hasattr(hidden_data_suffix, 'tmp_file'): | |
| 96 | - hidden_data_suffix = np.fromfile(hidden_data_suffix.tmp_file, | |
| 97 | - np.uint8) | |
| 98 | - else: | |
| 99 | - hidden_data_suffix = np.fromfile(src_hidden, np.uint8) | |
| 100 | - | |
| 101 | - hid_data = np.append(np.zeros(4, np.uint8), hidden_data_suffix) | |
| 102 | - | |
| 103 | - # write out the size of the hidden data | |
| 104 | - size_hd = np.size(hid_data) - 4 | |
| 105 | - | |
| 106 | - for i in xrange(4): | |
| 107 | - hid_data[i] = size_hd % 256 | |
| 108 | - size_hd /= 256 | |
| 109 | - | |
| 110 | - crypto = self.core.crypto_manager.get_instance() | |
| 111 | - hid_data = crypto.encrypt(hid_data) | |
| 112 | - self.ui.display_status('Encryption of data: %s' % crypto.info()) | |
| 113 | - | |
| 114 | - if np.size(hid_data) * (8 / word_size) > np.size(cov_data): | |
| 115 | - raise Exception("Cover image is too small to embed data. Try " + | |
| 116 | - "increasing the word size or choosing a larger " + | |
| 117 | - "cover image.") | |
| 118 | - | |
| 119 | - # converting hid_data to a sequence of words whose length is specified | |
| 120 | - # by word_size | |
| 121 | - hid_data = self.lookup_tab.to_word_sequence(hid_data, word_size) | |
| 122 | - self.ui.set_progress(2) | |
| 123 | - embedded = cov_data[:np.size(hid_data)] | |
| 124 | - self.ui.set_progress(5) | |
| 125 | - embedded = (embedded & ((0xff >> word_size) | |
| 126 | - << word_size)) | hid_data | |
| 127 | - self.ui.set_progress(95) | |
| 128 | - cov_data[:np.size(hid_data)] = embedded | |
| 129 | - self.ui.set_progress(98) | |
| 130 | - cov_data = cov_data.reshape(orig_shape) | |
| 131 | - cov_img = sp.misc.toimage(cov_data) | |
| 132 | - | |
| 133 | - if self.core.media_manager.is_media_key(tgt_stego): | |
| 134 | - self.core.media_manager.put_media(tgt_stego, cov_img) | |
| 135 | - else: | |
| 136 | - cov_img.save(tgt_stego) | |
| 137 | - | |
| 138 | - size_embedded = np.size(hid_data) / (8 / word_size) | |
| 139 | - self.ui.set_progress(99) | |
| 140 | - self._display_stats("embedded", np.size(cov_data), | |
| 141 | - size_embedded, time.time() - t0) | |
| 142 | - | |
| 143 | - @describe_annotate_convert((None, None, ident), | |
| 144 | - ("stego image", ImagePath, str), | |
| 145 | - ("hidden data", NewFilePath, str), | |
| 146 | - ("word size", [1, 2, 4, 8], int)) | |
| 147 | - def extract_raw_data(self, src_steg, tgt_hidden, word_size): | |
| 148 | - """ <p>This method extracts secret data from a stego image. It is | |
| 149 | - (obviously) the inverse operation of embed_raw_data.</p> | |
| 150 | - | |
| 151 | - <p>Parameters: | |
| 152 | - <ol> | |
| 153 | - <li><pre>src_stego</pre> | |
| 154 | - A valid pathname to an image file which serves as stego image.</li> | |
| 155 | - | |
| 156 | - <li><pre>tgt_hidden</pre> | |
| 157 | - A pathname denoting where the extracted data should be saved to.</li> | |
| 158 | - | |
| 159 | - <li><pre>word_size</pre> | |
| 160 | - Number of overwritten bits when the data was embedded.</li> | |
| 161 | - </ol> | |
| 162 | - An exception is raised if supposed secret data is too large to fit in | |
| 163 | - the stego image.</p> | |
| 77 | + Just a small helper function to extract hidden data. | |
| 164 | 78 | """ |
| 165 | - t0 = time.time() | |
| 166 | - self.ui.set_progress(1) | |
| 167 | - word_size = int(word_size) | |
| 168 | - steg_img = self.core.media_manager.get_file(src_steg) | |
| 169 | - | |
| 170 | - if steg_img and isinstance(steg_img, Image.Image): | |
| 171 | - steg_data = sp.misc.fromimage(steg_img).ravel() | |
| 172 | - elif steg_img and hasattr(steg_img, "data"): | |
| 173 | - raise Exception("stego image must be an image, \ | |
| 174 | - not an arbitrary data file") | |
| 175 | - else: | |
| 176 | - steg_data = sp.misc.fromimage(Image.open(src_steg)).ravel() | |
| 177 | - | |
| 178 | - # recovering file size | |
| 179 | - header_size = 4 * (8 / word_size) | |
| 180 | - size_data = steg_data[:header_size] & (0xff >> (8 - word_size)) | |
| 181 | - | |
| 182 | - size_data = self.lookup_tab.to_byte_sequence(size_data, word_size) | |
| 183 | - crypto = self.core.crypto_manager.get_instance() | |
| 184 | - size_data = crypto.decrypt(size_data) | |
| 185 | - self.ui.display_status('Decryption of size data: %s' % crypto.info()) | |
| 186 | - | |
| 187 | - size_hd = 0 | |
| 188 | - for i in xrange(4): | |
| 189 | - size_hd += size_data[i] * 256 ** i | |
| 190 | - | |
| 191 | - raw_size = size_hd * (8 / word_size) | |
| 192 | - | |
| 193 | - if raw_size > np.size(steg_data): | |
| 194 | - raise Exception("Supposed secret data too large for stego image.") | |
| 195 | - self.ui.set_progress(5) | |
| 196 | - hid_data = steg_data[header_size:raw_size + header_size] \ | |
| 197 | - & (0xff >> (8 - word_size)) | |
| 198 | - self.ui.set_progress(90) | |
| 199 | - hid_data = self.lookup_tab.to_byte_sequence(hid_data, word_size) | |
| 200 | - self.ui.set_progress(95) | |
| 201 | - final_data = crypto.decrypt(hid_data) | |
| 202 | - self.ui.display_status('Decryption of data: %s' % crypto.info()) | |
| 203 | - | |
| 204 | - self._extract_to_image_pool(tgt_hidden, final_data) | |
| 205 | - | |
| 206 | - self.ui.set_progress(99) | |
| 207 | - self._display_stats("extracted", np.size(steg_data), | |
| 208 | - np.size(hid_data) + np.size(size_data), | |
| 209 | - time.time() - t0) | |
| 79 | + hid_data = np.zeros(num_bits, np.uint8) | |
| 80 | + j = 0 | |
| 81 | + cnt = -1 | |
| 82 | + for x in np.nditer(steg_data): | |
| 83 | + cnt = cnt + 1 | |
| 84 | + if x == 0 or x == 1 or cnt % 64 == 0: continue | |
| 85 | + if j >= num_bits: break | |
| 86 | + hid_data[j] = x & 1 | |
| 87 | + j = j + 1 | |
| 88 | + | |
| 89 | + return hid_data | |
| 210 | 90 | |
| 211 | 91 | def __str__(self): |
| 212 | 92 | return 'LSB' | ... | ... |