MPB finished. The next step is dataset.

Chunk
1 parent 26e2fe9f
Showing 15 changed files with 83 additions and 6 deletions Show diff stats
common.pyc
jpegObj/__init__.pyc
jpegObj/base.pyc
jpegObj/dct.pyc
msteg/StegBase.pyc
msteg/__init__.pyc
msteg/steganalysis/MPB.py
msteg/steganalysis/MPB.pyc
msteg/steganalysis/__init__.pyc
msteg/steganography/F3.pyc
msteg/steganography/F4.pyc
msteg/steganography/F5.pyc
msteg/steganography/LSB.pyc
msteg/steganography/__init__.pyc
test_jpeg.py
@@ -14,10 +14,12 @@ from common import *
 import csv
 import json
 import pickle
+import cv2
 from sklearn import svm
  
 base_dir = '/home/hadoop/data/HeadShoulder/'
  
+
 class MPB(StegBase):
     """
     Markov Process Based Steganalyasis Algo.
@@ -25,8 +27,10 @@ class MPB(StegBase):
  
     def __init__(self):
         StegBase.__init__(self, sample_key)
+        self.model = None
+        self.svm = None
  
-    def get_trans_prob_mat_orig(self, ciq, T=4):
+    def _get_trans_prob_mat_orig(self, ciq, T=4):
         """
         Original!
         Calculate Transition Probability Matrix.
@@ -88,7 +92,10 @@ class MPB(StegBase):
         :param T: signed integer, usually 1~7
         :return: TPM - 3-D tensor, numpy array of size (2*T+1, 2*T+1, 4)
         """
-        # return self.get_trans_prob_mat_orig(ciq, T)
+
+        # return self._get_trans_prob_mat_orig(ciq, T)
+
+
         # timer = Timer()
         ciq = np.absolute(ciq).clip(0, T)
         TPM = np.zeros((2 * T + 1, 2 * T + 1, 4), np.float64)
@@ -159,13 +166,23 @@ class MPB(StegBase):
  
         return TPM
  
-    def _load_dataset(self,list_file):
+    def load_dataset(self, mode, file):
+        if mode == 'local':
+            return self._load_dataset_from_local(file)
+        elif mode == 'remote' or mode == 'hbase':
+            return self._load_dataset_from_hbase(file)
+        else:
+            raise Exception("Unknown mode!")
+
+    def _load_dataset_from_local(self, list_file='images_map_Train.tsv'):
         """
         load jpeg dataset according to a file of file-list.
  
         :param list_file: a tsv file with each line for a jpeg file path
         :return:(X,Y) for SVM
         """
+        list_file = base_dir + list_file
+
         X = []
         Y = []
         dict_tagbuf = {}
@@ -194,13 +211,76 @@ class MPB(StegBase):
         return X, Y
  
  
+    def _load_dataset_from_hbase(self, table='ImgCV'):
+        pass
+
+
+    def _model_svm_train_sk(self, X, Y):
+        timer = Timer()
+        timer.mark()
+        lin_clf = svm.LinearSVC()
+        lin_clf.fit(X, Y)
+        with open('res/tmp.model', 'wb') as modelfile:
+            model = pickle.dump(lin_clf, modelfile)
+
+        timer.report()
+
+        self.svm = 'sk'
+        self.model = lin_clf
+
+        return lin_clf
+
+    def _model_svm_predict_sk(self, image, clf=None):
+        if clf is None:
+            if self.svm == 'sk' and self.model != None:
+                clf = self.model
+            else:
+                with open('res/tmp.model', 'rb') as modelfile:
+                    clf = pickle.load(modelfile)
+
+        im = jpegObj.Jpeg(image, key=sample_key)
+        ciq = im.coef_arrays[jpegObj.colorMap['Y']]
+        tpm = self.get_trans_prob_mat(ciq)
+
+        return clf.predict(tpm)
+
+
+    def _model_svm_train_cv(self, X, Y):
+        svm_params = dict(kernel_type=cv2.SVM_LINEAR,
+                          svm_type=cv2.SVM_C_SVC,
+                          C=2.67, gamma=5.383)
+
+        timer = Timer()
+        timer.mark()
+        svm = cv2.SVM()
+        svm.train(X, Y, params=svm_params)
+        svm.save('res/svm_data.model')
  
+        self.svm = 'cv'
+        self.model = svm
  
+        return svm
  
+    def _model_svm_predict_cv(self, image, svm=None):
+        if svm is None:
+            if self.svm == 'cv' and self.model != None:
+                clf = self.model
+            else:
+                svm = cv2.SVM()
+                svm.load('res/svm_data.model')
  
+        im = jpegObj.Jpeg(image, key=sample_key)
+        ciq = im.coef_arrays[jpegObj.colorMap['Y']]
+        tpm = self.get_trans_prob_mat(ciq)
  
+        return svm.predict(tpm)
  
+    def train_svm(self):
+        X, Y = self.load_dataset('local', 'images_map_Train.tsv')
+        return self._model_svm_train_sk(X, Y)
  
+    def predict_svm(self,image):
+        return self._model_svm_predict_sk(image)
  
  
  
@@ -22,9 +22,6 @@ sample_key = [46812L, 20559L, 31360L, 16681L, 27536L, 39553L, 5427L, 63029L, 565
               5908L, 59816L, 56765L]
  
  
-
-
-
 def test_setblocks():
     """
     wholewise
...	...	@@ -14,10 +14,12 @@ from common import *
14	14	import csv
15	15	import json
16	16	import pickle
	17	+import cv2
17	18	from sklearn import svm
18	19
19	20	base_dir = '/home/hadoop/data/HeadShoulder/'
20	21
	22	+
21	23	class MPB(StegBase):
22	24	"""
23	25	Markov Process Based Steganalyasis Algo.
...	...	@@ -25,8 +27,10 @@ class MPB(StegBase):
25	27
26	28	def __init__(self):
27	29	StegBase.__init__(self, sample_key)
	30	+ self.model = None
	31	+ self.svm = None
28	32
29		- def get_trans_prob_mat_orig(self, ciq, T=4):
	33	+ def _get_trans_prob_mat_orig(self, ciq, T=4):
30	34	"""
31	35	Original!
32	36	Calculate Transition Probability Matrix.
...	...	@@ -88,7 +92,10 @@ class MPB(StegBase):
88	92	:param T: signed integer, usually 1~7
89	93	:return: TPM - 3-D tensor, numpy array of size (2T+1, 2T+1, 4)
90	94	"""
91		- # return self.get_trans_prob_mat_orig(ciq, T)
	95	+
	96	+ # return self._get_trans_prob_mat_orig(ciq, T)
	97	+
	98	+
92	99	# timer = Timer()
93	100	ciq = np.absolute(ciq).clip(0, T)
94	101	TPM = np.zeros((2 * T + 1, 2 * T + 1, 4), np.float64)
...	...	@@ -159,13 +166,23 @@ class MPB(StegBase):
159	166
160	167	return TPM
161	168
162		- def _load_dataset(self,list_file):
	169	+ def load_dataset(self, mode, file):
	170	+ if mode == 'local':
	171	+ return self._load_dataset_from_local(file)
	172	+ elif mode == 'remote' or mode == 'hbase':
	173	+ return self._load_dataset_from_hbase(file)
	174	+ else:
	175	+ raise Exception("Unknown mode!")
	176	+
	177	+ def _load_dataset_from_local(self, list_file='images_map_Train.tsv'):
163	178	"""
164	179	load jpeg dataset according to a file of file-list.
165	180
166	181	:param list_file: a tsv file with each line for a jpeg file path
167	182	:return:(X,Y) for SVM
168	183	"""
	184	+ list_file = base_dir + list_file
	185	+
169	186	X = []
170	187	Y = []
171	188	dict_tagbuf = {}
...	...	@@ -194,13 +211,76 @@ class MPB(StegBase):
194	211	return X, Y
195	212
196	213
	214	+ def _load_dataset_from_hbase(self, table='ImgCV'):
	215	+ pass
	216	+
	217	+
	218	+ def _model_svm_train_sk(self, X, Y):
	219	+ timer = Timer()
	220	+ timer.mark()
	221	+ lin_clf = svm.LinearSVC()
	222	+ lin_clf.fit(X, Y)
	223	+ with open('res/tmp.model', 'wb') as modelfile:
	224	+ model = pickle.dump(lin_clf, modelfile)
	225	+
	226	+ timer.report()
	227	+
	228	+ self.svm = 'sk'
	229	+ self.model = lin_clf
	230	+
	231	+ return lin_clf
	232	+
	233	+ def _model_svm_predict_sk(self, image, clf=None):
	234	+ if clf is None:
	235	+ if self.svm == 'sk' and self.model != None:
	236	+ clf = self.model
	237	+ else:
	238	+ with open('res/tmp.model', 'rb') as modelfile:
	239	+ clf = pickle.load(modelfile)
	240	+
	241	+ im = jpegObj.Jpeg(image, key=sample_key)
	242	+ ciq = im.coef_arrays[jpegObj.colorMap['Y']]
	243	+ tpm = self.get_trans_prob_mat(ciq)
	244	+
	245	+ return clf.predict(tpm)
	246	+
	247	+
	248	+ def _model_svm_train_cv(self, X, Y):
	249	+ svm_params = dict(kernel_type=cv2.SVM_LINEAR,
	250	+ svm_type=cv2.SVM_C_SVC,
	251	+ C=2.67, gamma=5.383)
	252	+
	253	+ timer = Timer()
	254	+ timer.mark()
	255	+ svm = cv2.SVM()
	256	+ svm.train(X, Y, params=svm_params)
	257	+ svm.save('res/svm_data.model')
197	258
	259	+ self.svm = 'cv'
	260	+ self.model = svm
198	261
	262	+ return svm
199	263
	264	+ def _model_svm_predict_cv(self, image, svm=None):
	265	+ if svm is None:
	266	+ if self.svm == 'cv' and self.model != None:
	267	+ clf = self.model
	268	+ else:
	269	+ svm = cv2.SVM()
	270	+ svm.load('res/svm_data.model')
200	271
	272	+ im = jpegObj.Jpeg(image, key=sample_key)
	273	+ ciq = im.coef_arrays[jpegObj.colorMap['Y']]
	274	+ tpm = self.get_trans_prob_mat(ciq)
201	275
	276	+ return svm.predict(tpm)
202	277
	278	+ def train_svm(self):
	279	+ X, Y = self.load_dataset('local', 'images_map_Train.tsv')
	280	+ return self._model_svm_train_sk(X, Y)
203	281
	282	+ def predict_svm(self,image):
	283	+ return self._model_svm_predict_sk(image)
204	284
205	285
206	286
...	...
...	...	@@ -22,9 +22,6 @@ sample_key = [46812L, 20559L, 31360L, 16681L, 27536L, 39553L, 5427L, 63029L, 565
22	22	5908L, 59816L, 56765L]
23	23
24	24
25		-
26		-
27		-
28	25	def test_setblocks():
29	26	"""
30	27	wholewise
...	...