-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun.py
224 lines (187 loc) · 7.08 KB
/
run.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
import conseal as cl
from glob import glob
import jpeglib
import numpy as np
import os
import pandas as pd
from pathlib import Path
from PIL import Image
from tempfile import NamedTemporaryFile
from tqdm import tqdm
#
df = []
# for fname in glob('test/assets/cover/jpeg_75_gray/*.jpg'):
files = glob(str(Path(os.environ['DATA']) / 'data/alaska2/fabrika-2024-01-26/images_ahd/*.png'))[:1]
for fname in tqdm(files):
x0 = np.array(Image.open(fname).convert('L'))
rho_p1, rho_m1 = cl.hill.compute_cost_adjusted(x0)
(p_p1, p_m1), lbda = cl.simulate._ternary.probability(
(rho_p1, rho_m1),
.01, # distortion
n=x0.size,
sender=cl.DLS,
)
cost_hat = (np.sum(p_p1 * rho_p1) + np.sum(p_m1 * rho_m1)) / x0.size
H_hat = cl.tools.entropy(p_p1, p_m1) / x0.size
print(fname, rho_p1.shape, rho_m1.shape, cost_hat, H_hat)
exit()
def attack(y1, qt):
# cartesian calibration
with NamedTemporaryFile(suffix='jpeg') as tmp:
jpeglib.from_dct(Y=y1, qt=qt).write_dct(tmp.name)
x1 = jpeglib.read_spatial(tmp.name).spatial
x2 = x1[3:-5, 3:-5] # crop 4x4
jpeglib.from_spatial(x2).write_spatial(tmp.name, qt=jpeg0.qt)
y2 = jpeglib.read_dct(tmp.name).Y
# #
# h1_01, _ = np.histogram(y1[:, :, 0, 1].flatten(), 8, range=(-4, 4))
# h1_10, _ = np.histogram(y1[:, :, 1, 0].flatten(), 8, range=(-4, 4))
# h1_11, _ = np.histogram(y1[:, :, 1, 1].flatten(), 8, range=(-4, 4))
# h2_01, _ = np.histogram(y2[:, :, 0, 1].flatten(), 8, range=(-4, 4))
# h2_10, _ = np.histogram(y2[:, :, 1, 0].flatten(), 8, range=(-4, 4))
# h2_11, _ = np.histogram(y2[:, :, 1, 1].flatten(), 8, range=(-4, 4))
# #
# beta_01 = (h1_01[0+4] - h2_01[0+4]) / (h2_01[-1+4] + h2_01[1+4])
# beta_10 = (h1_10[0+4] - h2_10[0+4]) / (h2_10[-1+4] + h2_10[1+4])
# beta_11 = (h1_11[0+4] - h2_11[0+4]) / (h2_11[-1+4] + h2_11[1+4])
#
h1_01, _ = np.histogram(np.abs(y1[:, :, 0, 1]).flatten(), 3, range=(0, 3))
h1_10, _ = np.histogram(np.abs(y1[:, :, 1, 0]).flatten(), 3, range=(0, 3))
h1_11, _ = np.histogram(np.abs(y1[:, :, 1, 1]).flatten(), 3, range=(0, 3))
h2_01, _ = np.histogram(np.abs(y2[:, :, 0, 1]).flatten(), 3, range=(0, 3))
h2_10, _ = np.histogram(np.abs(y2[:, :, 1, 0]).flatten(), 3, range=(0, 3))
h2_11, _ = np.histogram(np.abs(y2[:, :, 1, 1]).flatten(), 3, range=(0, 3))
beta_01 = (h2_01[1] * (h1_01[0] - h2_01[0]) + (h1_01[1] - h2_01[1]) * (h2_01[2] - h2_01[1])) / (h2_01[1]**2 + (h2_01[2] - h2_01[1])**2)
beta_10 = (h2_10[1] * (h1_10[0] - h2_10[0]) + (h1_10[1] - h2_10[1]) * (h2_10[2] - h2_10[1])) / (h2_10[1]**2 + (h2_10[2] - h2_10[1])**2)
beta_11 = (h2_11[1] * (h1_11[0] - h2_11[0]) + (h1_11[1] - h2_11[1]) * (h2_11[2] - h2_11[1])) / (h2_11[1]**2 + (h2_11[2] - h2_11[1])**2)
#
beta_hat = np.mean([beta_01, beta_10, beta_11])
beta_hat = np.clip(beta_hat, 0, None)
return beta_hat
#
df = []
# for fname in glob('test/assets/cover/jpeg_75_gray/*.jpg'):
files = glob(str(Path(os.environ['DATA']) / 'data/alaska2/fabrika-2024-01-26/images_ahd/*.png'))[:200]
for fname in tqdm(files):
x0 = np.array(Image.open(fname))
with NamedTemporaryFile(suffix='jpeg') as tmp:
jpeglib.from_spatial(x0).write_spatial(tmp.name, qt=75)
jpeg0 = jpeglib.read_dct(tmp.name)
y0, qt0 = jpeg0.Y, jpeg0.qt[0]
#
for alpha in [.4]:
beta_hat = attack(y0, qt0)
#
y1_F5 = cl.F5.simulate_single_channel(y0=y0, alpha=alpha, seed=12345)
beta_F5 = (y0 != y1_F5).sum() / cl.tools.nzAC(y0)
beta_hat_F5 = attack(y1_F5, qt0)
#
y1_nsF5 = cl.nsF5.simulate_single_channel(y0=y0, alpha=alpha, seed=12345)
beta_nsF5 = (y0 != y1_nsF5).sum() / cl.tools.nzAC(y0)
beta_hat_nsF5 = attack(y1_nsF5, qt0)
# print(beta_hat, beta, '|', np.abs(beta - beta_hat))
# print('\n', np.abs(beta - beta_hat))
df.append({
'fname': Path(fname).name,
'alpha': alpha,
'beta_hat': beta_hat,
'beta_F5': beta_F5,
'beta_hat_F5': beta_hat_F5,
'beta_nsF5': beta_nsF5,
'beta_hat_nsF5': beta_hat_nsF5,
'mae0': np.abs(0 - beta_hat),
'mae1_F5': np.abs(beta_F5 - beta_hat_F5),
'mae1_nsF5': np.abs(beta_nsF5 - beta_hat_nsF5),
})
df = pd.DataFrame(df)
print(df)
print(df[['beta_hat', 'beta_F5', 'beta_hat_F5', 'beta_nsF5', 'beta_hat_nsF5', 'mae0', 'mae1_F5', 'mae1_nsF5']].mean())
# import numpy as np
# def soliton(
# m: int = 100,
# *,
# robust: bool = True,
# c: float = .1,
# delta: float = .5,
# ) -> np.ndarray:
# """
# :param m:
# :type m: int
# :param c:
# :type c: float
# :param delta:
# :type delta: float
# :return:
# :rtype: np.ndarray
# """
# # Soliton distribution
# i = np.arange(1, m+1)
# p = nu = np.concatenate([[1/m], 1/(i[1:]*i[:-1])])
# # robust Soliton distribution
# if robust:
# T = c * np.log(m / delta) * np.sqrt(m)
# mT = int(np.floor(m / T))
# # print(m, mT)
# assert m > mT
# tau = np.concatenate([
# T / (i[:mT-1] * m),
# [T * np.log(T / delta) / m],
# np.zeros(m - mT),
# ])
# p = nu + tau
# #
# p = (nu + tau)
# return p / np.sum(p)
# def generate_H(
# m: int,
# n: int = None,
# *,
# c: float = .1,
# delta: float = .5,
# seed: int = None,
# ) -> np.ndarray:
# """Generates parity-check matrix according to robust soliton distribution.
# :param m: number of rows/message bits
# :type m: int
# :param n: number of columns/cover elements
# :type n: int
# :param c: constant parameter
# :type c: float
# :param delta: failure probability
# :type delta: float
# :return: parity-check matrix
# :rtype: np.ndarray
# :Example:
# >>> # TODO
# """
# if n is None:
# n = 2**m-1
# # Get the robust Soliton distribution
# rsd = soliton(m, c=c, delta=delta, robust=True)
# # Sample column weights w[1], ..., w[n] from the RSD
# rng = np.random.default_rng(seed)
# column_weights = rng.choice(m, size=n, p=rsd) + 1
# # Generate columns of H
# H = np.zeros((m, n), dtype=int)
# for j in range(n):
# # Create a column with column_weights[j] ones
# ones_positions = rng.choice(m, size=column_weights[j], replace=False)
# H[ones_positions, j] = 1
# return H
# # Parameters
# m = 100 # Number of rows
# # n = 10000 # Number of columns
# n = 1000
# c = .1 # Constant parameter
# delta = .05 # Failure probability
# # Generate the parity-check matrix
# H = generate_H(m, n=n, delta=delta, c=c, seed=12345)
# print("Generated Parity-Check Matrix:\n", H)
# #
# p_empirical = np.histogram(np.sum(H, axis=0), bins=m-1, range=(1, m), density=True)[0]
# p_theoretical = soliton(m=m, c=c, delta=delta)
# import matplotlib.pyplot as plt
# fig, ax = plt.subplots()
# ax.plot(range(50), p_theoretical[:50], label="Theoretical RSD", linewidth=1)
# ax.plot(range(50), p_empirical[:50], label="Empirical RSD", linewidth=1, linestyle='dotted')
# plt.show()