def _detect_e4s_srm(files=None, file_path=None, model_file="e4s_srm_bossbase_lsbm0.10_gs.model", imgs=None): # 1 = clean, 0: stego from aletheialib import stegosim, feaext, models extractor = "srm" clf = models.Ensemble4Stego() clf.load("{}{}".format(alethia_model_paths, model_file)) stat_score = [] if not files and file_path: files = [file_path] for i, f in enumerate(files): img = array_to_img(imgs[i]) if imgs else None X = feaext.extractor_fn(extractor)(f, im=img) X = X.reshape((1, X.shape[0])) p = clf.predict(X) if p[0] == 0: stat_score.append(1) else: stat_score.append(0) return np.array(stat_score)
def main(): #" - aump: Adaptive Steganalysis Attack.\n" \ attacks_doc="\n" \ " Statistical attacks:\n" \ " - sp: Sample Pairs Analysis (Octave vesion).\n" \ " - ws: Weighted Stego Attack.\n" \ " - triples: Triples Attack.\n" \ " - spa: Sample Pairs Analysis.\n" \ " - rs: RS attack.\n" \ " - calibration: Calibration attack to JPEG images." embsim_doc="\n" \ " Embedding simulators:\n" \ " - lsbr-sim: Embedding using LSB replacement simulator.\n" \ " - lsbm-sim: Embedding using LSB matching simulator.\n" \ " - hugo-sim: Embedding using HUGO simulator.\n" \ " - wow-sim: Embedding using WOW simulator.\n" \ " - s-uniward-sim: Embedding using S-UNIWARD simulator.\n" \ " - j-uniward-sim: Embedding using J-UNIWARD simulator.\n" \ " - j-uniward-color-sim: Embedding using J-UNIWARD color simulator.\n" \ " - hill-sim: Embedding using HILL simulator.\n" \ " - ebs-sim: Embedding using EBS simulator.\n" \ " - ebs-color-sim: Embedding using EBS color simulator.\n" \ " - ued-sim: Embedding using UED simulator.\n" \ " - ued-color-sim: Embedding using UED color simulator.\n" \ " - nsf5-sim: Embedding using nsF5 simulator.\n" \ " - nsf5-color-sim: Embedding using nsF5 color simulator." model_doc="\n" \ " Model training:\n" \ " - esvm: Ensemble of Support Vector Machines.\n" \ " - e4s: Ensemble Classifiers for Steganalysis.\n" \ " - srnet: Steganalysis Residual Network." mldetect_doc="\n" \ " ML-based detectors:\n" \ " - esvm-predict: Predict using eSVM.\n" \ " - e4s-predict: Predict using EC.\n" \ " - srnet-predict: Predict using SRNet.\n" \ " - srnet-err: Score error using SRNet.\n" \ " - srnet-err-icd: Score ICD error using SRNet." feaextract_doc="\n" \ " Feature extractors:\n" \ " - srm: Full Spatial Rich Models.\n" \ " - hill-maxsrm: Selection-Channel-Aware Spatial Rich Models for HILL.\n" \ " - srmq1: Spatial Rich Models with fixed quantization q=1c.\n" \ " - scrmq1: Spatial Color Rich Models with fixed quantization q=1c.\n" \ " - gfr: JPEG steganalysis with 2D Gabor Filters." auto_doc="\n" \ " Unsupervised attacks:\n" \ " - ats: Artificial Training Sets." tools_doc="\n" \ " Tools:\n" \ " - brute-force: Brute force attack using a list of passwords.\n" \ " - hpf: High-pass filter.\n" \ " - print-diffs: Differences between two images.\n" \ " - print-dct-diffs: Differences between the DCT coefficients of two JPEG images.\n" \ " - rm-alpha: Opacity of the alpha channel to 255.\n" \ " - prep-ml-exp: Prepare an experiment for testing ML tools." if len(sys.argv)<2: print(sys.argv[0], "<command>\n") print("COMMANDS:") print(attacks_doc) print(mldetect_doc) print(feaextract_doc) print(embsim_doc) print(model_doc) print(auto_doc) print(tools_doc) print("\n") sys.exit(0) if False: pass # -- ATTACKS -- # {{{ sp elif sys.argv[1]=="sp": if len(sys.argv)!=3: print(sys.argv[0], "spa <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) threshold=0.05 path = utils.absolute_path(sys.argv[2]) im=Image.open(path) if im.mode in ['RGB', 'RGBA', 'RGBX']: alpha_R = _attack('SP', path, params={"channel":1})["data"][0][0] alpha_G = _attack('SP', path, params={"channel":2})["data"][0][0] alpha_B = _attack('SP', path, params={"channel":3})["data"][0][0] if alpha_R<threshold and alpha_G<threshold and alpha_B<threshold: print("No hidden data found") if alpha_R>=threshold: print("Hiden data found in channel R", alpha_R) if alpha_G>=threshold: print("Hiden data found in channel G", alpha_G) if alpha_B>=threshold: print("Hiden data found in channel B", alpha_B) else: alpha = _attack('SP', path, params={"channel":1})["data"][0][0] if alpha>=threshold: print("Hiden data found", alpha) else: print("No hidden data found") sys.exit(0) # }}} # {{{ ws elif sys.argv[1]=="ws": if len(sys.argv)!=3: print(sys.argv[0], "ws <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) threshold=0.05 path = utils.absolute_path(sys.argv[2]) im=Image.open(path) if im.mode in ['RGB', 'RGBA', 'RGBX']: alpha_R = _attack('WS', path, params={"channel":1})["data"][0][0] alpha_G = _attack('WS', path, params={"channel":2})["data"][0][0] alpha_B = _attack('WS', path, params={"channel":3})["data"][0][0] if alpha_R<threshold and alpha_G<threshold and alpha_B<threshold: print("No hidden data found") if alpha_R>=threshold: print("Hiden data found in channel R", alpha_R) if alpha_G>=threshold: print("Hiden data found in channel G", alpha_G) if alpha_B>=threshold: print("Hiden data found in channel B", alpha_B) else: alpha = _attack('WS', path, params={"channel":1})["data"][0][0] if alpha>=threshold: print("Hiden data found", alpha) else: print("No hidden data found") sys.exit(0) # }}} # {{{ triples elif sys.argv[1]=="triples": if len(sys.argv)!=3: print(sys.argv[0], "triples <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) threshold=0.05 path = utils.absolute_path(sys.argv[2]) im=Image.open(path) if im.mode in ['RGB', 'RGBA', 'RGBX']: alpha_R = _attack('TRIPLES', path, params={"channel":1})["data"][0][0] alpha_G = _attack('TRIPLES', path, params={"channel":2})["data"][0][0] alpha_B = _attack('TRIPLES', path, params={"channel":3})["data"][0][0] if alpha_R<threshold and alpha_G<threshold and alpha_B<threshold: print("No hidden data found") if alpha_R>=threshold: print("Hiden data found in channel R", alpha_R) if alpha_G>=threshold: print("Hiden data found in channel G", alpha_G) if alpha_B>=threshold: print("Hiden data found in channel B", alpha_B) else: alpha = _attack('TRIPLES', path, params={"channel":1})["data"][0][0] if alpha>=threshold: print("Hiden data found", alpha) else: print("No hidden data found") sys.exit(0) # }}} # {{{ aump elif sys.argv[1]=="aump": if len(sys.argv)!=3: print(sys.argv[0], "aump <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) path = utils.absolute_path(sys.argv[2]) im=Image.open(path) if im.mode in ['RGB', 'RGBA', 'RGBX']: beta_R = _attack('AUMP', path, params={"channel":1})["data"][0][0] beta_G = _attack('AUMP', path, params={"channel":2})["data"][0][0] beta_B = _attack('AUMP', path, params={"channel":3})["data"][0][0] # XXX: How to use beta? else: beta = _attack('AUMP', path, params={"channel":1})["data"][0][0] # XXX: How to use beta? sys.exit(0) # }}} # {{{ spa elif sys.argv[1]=="spa": if len(sys.argv)!=3: print(sys.argv[0], "spa <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) threshold=0.05 I = imread(sys.argv[2]) if len(I.shape)==2: bitrate=attacks.spa_image(I, None) if bitrate<threshold: print("No hidden data found") else: print("Hiden data found"), bitrate else: bitrate_R=attacks.spa_image(I, 0) bitrate_G=attacks.spa_image(I, 1) bitrate_B=attacks.spa_image(I, 2) if bitrate_R<threshold and bitrate_G<threshold and bitrate_B<threshold: print("No hidden data found") sys.exit(0) if bitrate_R>=threshold: print("Hiden data found in channel R", bitrate_R) if bitrate_G>=threshold: print("Hiden data found in channel G", bitrate_G) if bitrate_B>=threshold: print("Hiden data found in channel B", bitrate_B) sys.exit(0) # }}} # {{{ rs elif sys.argv[1]=="rs": if len(sys.argv)!=3: print(sys.argv[0], "spa <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) threshold=0.05 I = imread(sys.argv[2]) if len(I.shape)==2: bitrate=attacks.rs(sys.argv[2], None) if bitrate<threshold: print("No hidden data found") else: print("Hiden data found", bitrate) else: bitrate_R=attacks.rs(sys.argv[2], 0) bitrate_G=attacks.rs(sys.argv[2], 1) bitrate_B=attacks.rs(sys.argv[2], 2) if bitrate_R<threshold and bitrate_G<threshold and bitrate_B<threshold: print("No hidden data found") sys.exit(0) if bitrate_R>=threshold: print("Hiden data found in channel R", bitrate_R) if bitrate_G>=threshold: print("Hiden data found in channel G", bitrate_G) if bitrate_B>=threshold: print("Hiden data found in channel B", bitrate_B) sys.exit(0) # }}} # {{{ calibration elif sys.argv[1]=="calibration": if len(sys.argv)!=3: print(sys.argv[0], "calibration <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) fn = utils.absolute_path(sys.argv[2]) attacks.calibration(fn) # }}} # -- ML-BASED DETECTORS -- # {{{ esvm-predict elif sys.argv[1]=="esvm-predict": if len(sys.argv)!=5: print(sys.argv[0], "esvm-predict <model-file> <feature-extractor> <image/dir>") print(feaextract_doc) sys.exit(0) model_file=sys.argv[2] extractor=sys.argv[3] path=utils.absolute_path(sys.argv[4]) files=[] if os.path.isdir(path): for dirpath,_,filenames in os.walk(path): for f in filenames: path=os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files=[path] clf=pickle.load(open(model_file, "r")) for f in files: X = feaext.extractor_fn(extractor)(f) X = X.reshape((1, X.shape[0])) p = clf.predict_proba(X) print(p) if p[0][0] > 0.5: print(os.path.basename(f), "Cover, probability:", p[0][0]) else: print(os.path.basename(f), "Stego, probability:", p[0][1]) # }}} # {{{ e4s-predict elif sys.argv[1]=="e4s-predict": from aletheialib import models if len(sys.argv)!=5: print(sys.argv[0], "e4s-predict <model-file> <feature-extractor> <image/dir>\n") print("") print(feaextract_doc) print("") sys.exit(0) model_file=sys.argv[2] extractor=sys.argv[3] path=utils.absolute_path(sys.argv[4]) files=[] if os.path.isdir(path): for dirpath,_,filenames in os.walk(path): for f in filenames: path=os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files=[path] from aletheialib import models clf=models.Ensemble4Stego() clf.load(model_file) for f in files: X = feaext.extractor_fn(extractor)(f) X = X.reshape((1, X.shape[0])) p = clf.predict(X) if p[0] == 0: print(os.path.basename(f), "Cover") else: print(os.path.basename(f), "Stego") # }}} # {{{ srnet-predict elif sys.argv[1]=="srnet-predict": from aletheialib import models if len(sys.argv)<4: print(sys.argv[0], "srnet-predict <model dir> <image/dir> [dev]\n") print(" dev: Device: GPU Id or 'CPU' (default='CPU')") print("") sys.exit(0) model_dir=sys.argv[2] path=utils.absolute_path(sys.argv[3]) if len(sys.argv)<4: dev_id = "CPU" print("'dev' not provided, using:", dev_id) else: dev_id = sys.argv[4] files=[] if os.path.isdir(path): for dirpath,_,filenames in os.walk(path): for f in filenames: path=os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files=[path] from aletheialib import models models.nn_configure_device(dev_id) pred = models.nn_predict(models.SRNet, files, model_dir, batch_size=20) #print(pred) for i in range(len(files)): if pred[i] == 0: print(os.path.basename(files[i]), "Cover") else: print(os.path.basename(files[i]), "Stego") # }}} # {{{ srnet-score elif sys.argv[1]=="srnet-err": from aletheialib import models if len(sys.argv)<4: print(sys.argv[0], "srnet-score <model dir> <cover dir> <stego dir> [dev]\n") print(" dev: Device: GPU Id or 'CPU' (default='CPU')") print("") sys.exit(0) model_dir=sys.argv[2] cover_path=utils.absolute_path(sys.argv[3]) stego_path=utils.absolute_path(sys.argv[4]) if len(sys.argv)<5: dev_id = "CPU" print("'dev' not provided, using:", dev_id) else: dev_id = sys.argv[5] path = cover_path files=[] if os.path.isdir(path): for dirpath,_,filenames in os.walk(path): for f in filenames: path=os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files=[path] cover_files = files path = stego_path files=[] if os.path.isdir(path): for dirpath,_,filenames in os.walk(path): for f in filenames: path=os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files=[path] stego_files = files from aletheialib import models models.nn_configure_device(dev_id) cover_pred = models.nn_predict(models.SRNet, cover_files, model_dir, batch_size=20) stego_pred = models.nn_predict(models.SRNet, stego_files, model_dir, batch_size=20) ok = 0 for i in range(len(cover_files)): if cover_pred[i] == 0: ok += 1 for i in range(len(stego_files)): if stego_pred[i] == 1: ok += 1 print("err:", 1-round(float(ok)/(len(cover_files)+len(stego_files)), 4)) # }}} # {{{ srnet-score elif sys.argv[1]=="srnet-err-icd": from aletheialib import models if len(sys.argv)<6: print(sys.argv[0], "srnet-score-icd <A model dir> <B model dir> <A cover dir> <A stego dir> <B cover dir> <B stego dir> [dev]\n") print(" dev: Device: GPU Id or 'CPU' (default='CPU')") print("") sys.exit(0) A_model_dir=sys.argv[2] B_model_dir=sys.argv[3] A_cover_path=utils.absolute_path(sys.argv[4]) A_stego_path=utils.absolute_path(sys.argv[5]) B_cover_path=utils.absolute_path(sys.argv[6]) B_stego_path=utils.absolute_path(sys.argv[7]) if len(sys.argv)<8: dev_id = "CPU" print("'dev' not provided, using:", dev_id) else: dev_id = sys.argv[8] def get_files(path): files=[] if os.path.isdir(path): for dirpath,_,filenames in os.walk(path): for f in filenames: path=os.path.abspath(os.path.join(dirpath, f)) if os.path.isdir(path) or not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files=[path] return files A_cover_files = get_files(A_cover_path) A_stego_files = get_files(A_stego_path) B_cover_files = get_files(B_cover_path) B_stego_files = get_files(B_stego_path) from aletheialib import models models.nn_configure_device(dev_id) p_aa_cover = models.nn_predict(models.SRNet, A_cover_files, A_model_dir, batch_size=20) p_aa_stego = models.nn_predict(models.SRNet, A_stego_files, A_model_dir, batch_size=20) p_aa = p_aa_cover.tolist() + p_aa_stego.tolist() p_bb_cover = models.nn_predict(models.SRNet, B_cover_files, B_model_dir, batch_size=20) p_bb_stego = models.nn_predict(models.SRNet, B_stego_files, B_model_dir, batch_size=20) p_bb = p_bb_cover.tolist() + p_bb_stego.tolist() p_ab_cover = models.nn_predict(models.SRNet, B_cover_files, A_model_dir, batch_size=20) p_ab_stego = models.nn_predict(models.SRNet, B_stego_files, A_model_dir, batch_size=20) p_ab = p_ab_cover.tolist() + p_ab_stego.tolist() p_ba_cover = models.nn_predict(models.SRNet, A_cover_files, B_model_dir, batch_size=20) p_ba_stego = models.nn_predict(models.SRNet, A_stego_files, B_model_dir, batch_size=20) p_ba = p_ba_cover.tolist() + p_ba_stego.tolist() ok = 0 for i in range(len(A_cover_files)): if p_aa_cover[i] == 0: ok += 1 for i in range(len(A_stego_files)): if p_aa_stego[i] == 1: ok += 1 score = round(float(ok)/(len(A_cover_files)+len(A_stego_files)), 4) pred, inc, xaS = inconsistencies.AB_predict(p_aa, p_bb, p_ab, p_ba) print("err:", round(1-score,2) , ", err pred:", round(float(inc)/(2*len(pred)),2), ", xa:", xaS) # }}} # -- FEATURE EXTRACTORS -- # {{{ srm elif sys.argv[1]=="srm": if len(sys.argv)!=4: print(sys.argv[0], "srm <image/dir> <output-file>\n") sys.exit(0) image_path=sys.argv[2] ofile=sys.argv[3] feaext.extract_features(feaext.SRM_extract, image_path, ofile) # }}} # {{{ srmq1 elif sys.argv[1]=="srmq1": if len(sys.argv)!=4: print(sys.argv[0], "srmq1 <image/dir> <output-file>\n") sys.exit(0) image_path=sys.argv[2] ofile=sys.argv[3] feaext.extract_features(feaext.SRMQ1_extract, image_path, ofile) # }}} # {{{ scrmq1 elif sys.argv[1]=="scrmq1": if len(sys.argv)!=4: print(sys.argv[0], "scrmq1 <image/dir> <output-file>\n") sys.exit(0) image_path=sys.argv[2] ofile=sys.argv[3] feaext.extract_features(feaext.SCRMQ1_extract, image_path, ofile) # }}} # {{{ gfr elif sys.argv[1]=="gfr": if len(sys.argv)<4: print(sys.argv[0], "gfr <image/dir> <output-file> [quality] [rotations]\n") sys.exit(0) image_path=sys.argv[2] ofile=sys.argv[3] if len(sys.argv)<5: quality = "auto" print("JPEG quality not provided, using detection via 'identify'") else: quality = sys.argv[4] if len(sys.argv)<6: rotations = 32 print("Number of rotations for Gabor kernel no provided, using:", \ rotations) else: rotations = sys.argv[6] params = { "quality": quality, "rotations": rotations } feaext.extract_features(feaext.GFR_extract, image_path, ofile, params) # }}} # {{{ dctr elif sys.argv[1]=="dctr": if len(sys.argv)<4: print(sys.argv[0], "dctr <image/dir> <output-file> [quality]\n") sys.exit(0) image_path=sys.argv[2] ofile=sys.argv[3] if len(sys.argv)<5: quality = "auto" print("JPEG quality not provided, using detection via 'identify'") else: quality = sys.argv[4] params = { "quality": quality, } feaext.extract_features(feaext.DCTR_extract, image_path, ofile, params) # }}} # {{{ hill-sigma-spam-psrm elif sys.argv[1]=="hill-sigma-spam-psrm": if len(sys.argv)!=4: print(sys.argv[0], "hill-sigma-spam-psrm <image/dir> <output-file>\n") sys.exit(0) image_path=sys.argv[2] ofile=sys.argv[3] feaext.extract_features(feaext.HILL_sigma_spam_PSRM_extract, image_path, ofile) # }}} # {{{ hill-maxsrm elif sys.argv[1]=="hill-maxsrm": if len(sys.argv)!=4: print(sys.argv[0], "hill-maxsrm <image/dir> <output-file>\n") sys.exit(0) image_path=sys.argv[2] ofile=sys.argv[3] feaext.extract_features(feaext.HILL_MAXSRM_extract, image_path, ofile) # }}} # -- EMBEDDING SIMULATORS -- # {{{ lsbr-sim elif sys.argv[1]=="lsbr-sim": if len(sys.argv)!=5: print(sys.argv[0], "lsbr-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.lsbr, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ lsbm-sim elif sys.argv[1]=="lsbm-sim": if len(sys.argv)!=5: print(sys.argv[0], "lsbm-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.lsbm, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ hugo-sim elif sys.argv[1]=="hugo-sim": if len(sys.argv)!=5: print(sys.argv[0], "hugo-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.hugo, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ wow-sim elif sys.argv[1]=="wow-sim": if len(sys.argv)!=5: print(sys.argv[0], "wow-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.wow, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ s-uniward-sim elif sys.argv[1]=="s-uniward-sim": if len(sys.argv)!=5: print(sys.argv[0], "s-uniward-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.s_uniward, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ hill-sim elif sys.argv[1]=="hill-sim": if len(sys.argv)!=5: print(sys.argv[0], "hill-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.hill, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ j-uniward-sim elif sys.argv[1]=="j-uniward-sim": if len(sys.argv)!=5: print(sys.argv[0], "j-uniward-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.j_uniward, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ j-uniward-color-sim elif sys.argv[1]=="j-uniward-color-sim": if len(sys.argv)!=5: print(sys.argv[0], "j-uniward-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.j_uniward_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ebs-sim elif sys.argv[1]=="ebs-sim": if len(sys.argv)!=5: print(sys.argv[0], "ebs-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ebs, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ebs-color-sim elif sys.argv[1]=="ebs-color-sim": if len(sys.argv)!=5: print(sys.argv[0], "ebs-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ebs_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ued-sim elif sys.argv[1]=="ued-sim": if len(sys.argv)!=5: print(sys.argv[0], "ued-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ued, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ued-color-sim elif sys.argv[1]=="ued-color-sim": if len(sys.argv)!=5: print(sys.argv[0], "ued-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ued_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ nsf5-sim elif sys.argv[1]=="nsf5-sim": if len(sys.argv)!=5: print(sys.argv[0], "nsf5-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.nsf5, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ nsf5-color-sim elif sys.argv[1]=="nsf5-color-sim": if len(sys.argv)!=5: print(sys.argv[0], "nsf5-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.nsf5_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ experimental-sim elif sys.argv[1]=="experimental-sim": if len(sys.argv)!=5: print(sys.argv[0], "experimental-sim <image/dir> <payload> <output-dir>") print("NOTE: Please, put your EXPERIMENTAL.m file into external/octave\n") sys.exit(0) stegosim.embed_message(stegosim.experimental, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # -- MODEL TRAINING -- # {{{ esvm elif sys.argv[1]=="esvm": from aletheialib import models if len(sys.argv)!=5: print(sys.argv[0], "esvm <cover-fea> <stego-fea> <model-file>\n") sys.exit(0) from sklearn.model_selection import train_test_split cover_fea=sys.argv[2] stego_fea=sys.argv[3] model_file=utils.absolute_path(sys.argv[4]) X_cover = pandas.read_csv(cover_fea, delimiter = " ").values X_stego = pandas.read_csv(stego_fea, delimiter = " ").values #X_cover=numpy.loadtxt(cover_fea) #X_stego=numpy.loadtxt(stego_fea) X=numpy.vstack((X_cover, X_stego)) y=numpy.hstack(([0]*len(X_cover), [1]*len(X_stego))) X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.10) from aletheialib import models clf=models.EnsembleSVM() clf.fit(X_train, y_train) val_score=clf.score(X_val, y_val) pickle.dump(clf, open(model_file, "wb")) print("Validation score:", val_score) # }}} # {{{ e4s elif sys.argv[1]=="e4s": from aletheialib import models if len(sys.argv)!=5: print(sys.argv[0], "e4s <cover-fea> <stego-fea> <model-file>\n") sys.exit(0) from sklearn.model_selection import train_test_split cover_fea=sys.argv[2] stego_fea=sys.argv[3] model_file=utils.absolute_path(sys.argv[4]) X_cover = pandas.read_csv(cover_fea, delimiter = " ").values X_stego = pandas.read_csv(stego_fea, delimiter = " ").values #X_cover=numpy.loadtxt(cover_fea) #X_stego=numpy.loadtxt(stego_fea) X=numpy.vstack((X_cover, X_stego)) y=numpy.hstack(([0]*len(X_cover), [1]*len(X_stego))) X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.10) clf=models.Ensemble4Stego() clf.fit(X_train, y_train) val_score=clf.score(X_val, y_val) clf.save(model_file) print("Validation score:", val_score) # }}} # {{{ srnet elif sys.argv[1]=="srnet": from aletheialib import models if len(sys.argv)<5: print(sys.argv[0], "srnet <cover-dir> <stego-dir> <model-name> [dev] [max_iter] [ES] [valsz] [logdir]\n") print(" dev: Device: GPU Id or 'CPU' (default='CPU')") print(" max_iter: Number of iterations (default=1000000)") print(" ES: early stopping iterations x1000 (default=100)") print(" valsz: Size of validation set. (default=0.1%)") print(" logdir: Log directory. (default=log)") print("") sys.exit(0) cover_dir=sys.argv[2] stego_dir=sys.argv[3] model_name=sys.argv[4] if len(sys.argv)<6: dev_id = "CPU" print("'dev' not provided, using:", dev_id) else: dev_id = sys.argv[5] if len(sys.argv)<7: max_iter = 1000000 print("'max_iter' not provided, using:", max_iter) else: max_iter = int(sys.argv[6]) if len(sys.argv)<8: early_stopping = 100 print("'ES' not provided, using:", early_stopping) else: early_stopping = int(sys.argv[7]) if len(sys.argv)<9: val_size = 0.1 print("'valsz' not provided, using:", val_size) else: val_size = float(sys.argv[8]) if len(sys.argv)<10: log_dir = 'log' print("'logdir' not provided, using:", log_dir) else: log_dir = sys.argv[9] if dev_id == "CPU": print("Running with CPU. It could be very slow!") from aletheialib import models models.nn_configure_device(dev_id) from sklearn.model_selection import train_test_split cover_files = sorted(glob.glob(os.path.join(cover_dir, '*'))) stego_files = sorted(glob.glob(os.path.join(stego_dir, '*'))) train_cover_files, valid_cover_files, train_stego_files, valid_stego_files = \ train_test_split(cover_files, stego_files, test_size=val_size, random_state=0) print("Using", len(train_cover_files)*2, "samples for training and", len(valid_cover_files)*2, "for validation.") output_dir = os.path.join(log_dir, 'output') checkpoint_dir = os.path.join(log_dir, 'checkpoint') for d in [output_dir, checkpoint_dir]: try: os.makedirs(d) except: pass data = (train_cover_files, train_stego_files, valid_cover_files, valid_stego_files) models.nn_fit(models.SRNet, data, model_name, log_path=output_dir, load_checkpoint=model_name, checkpoint_path=checkpoint_dir, batch_size=20, optimizer=models.AdamaxOptimizer(0.001), max_iter=max_iter, early_stopping=early_stopping, valid_interval=1000) # }}} # -- AUTOMATED ATTACKS -- # {{{ ats elif sys.argv[1]=="ats": from aletheialib import models if len(sys.argv) not in [5, 6]: print(sys.argv[0], "ats <embed-sim> <payload> <fea-extract> <images>") print(sys.argv[0], "ats <custom command> <fea-extract> <images>\n") print(embsim_doc) print("") print(feaextract_doc) print("") print("Examples:") print(sys.argv[0], "ats hill-sim 0.40 srm image_dir/") print(sys.argv[0], "ats 'steghide embed -cf <IMAGE> -ef secret.txt -p mypass' srm image_dir/\n") sys.exit(0) embed_fn_saving=False if len(sys.argv) == 6: emb_sim=sys.argv[2] payload=sys.argv[3] feaextract=sys.argv[4] A_dir=sys.argv[5] fn_sim=stegosim.embedding_fn(emb_sim) fn_feaextract=feaext.extractor_fn(feaextract) if emb_sim in ["j-uniward-sim", "j-uniward-color-sim", "ued-sim", "ued-color-sim", "ebs-sim", "ebs-color-sim", "nsf5-sim", "nsf5-color-sim"]: embed_fn_saving = True else: print("custom command") payload=sys.argv[2] # uggly hack feaextract=sys.argv[3] A_dir=sys.argv[4] fn_sim=stegosim.custom embed_fn_saving = True fn_feaextract=feaext.extractor_fn(feaextract) B_dir=tempfile.mkdtemp() C_dir=tempfile.mkdtemp() stegosim.embed_message(fn_sim, A_dir, payload, B_dir, embed_fn_saving=embed_fn_saving) stegosim.embed_message(fn_sim, B_dir, payload, C_dir, embed_fn_saving=embed_fn_saving) fea_dir=tempfile.mkdtemp() A_fea=os.path.join(fea_dir, "A.fea") C_fea=os.path.join(fea_dir, "C.fea") feaext.extract_features(fn_feaextract, A_dir, A_fea) feaext.extract_features(fn_feaextract, C_dir, C_fea) A = pandas.read_csv(A_fea, delimiter = " ").values C = pandas.read_csv(C_fea, delimiter = " ").values X=numpy.vstack((A, C)) y=numpy.hstack(([0]*len(A), [1]*len(C))) from aletheialib import models clf=models.Ensemble4Stego() clf.fit(X, y) files=[] for dirpath,_,filenames in os.walk(B_dir): for f in filenames: path=os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, this is not a valid image: ", f) else: files.append(path) for f in files: B = fn_feaextract(f) B = B.reshape((1, B.shape[0])) p = clf.predict(B) if p[0] == 0: print(os.path.basename(f), "Cover") else: print(os.path.basename(f), "Stego") shutil.rmtree(B_dir) shutil.rmtree(C_dir) shutil.rmtree(fea_dir) # }}} # -- NAIVE ATTACKS -- # {{{ brute-force elif sys.argv[1]=="brute-force": if len(sys.argv)!=4: print(sys.argv[0], "brute-force <unhide command> <passw file>\n") print("Example:") print(sys.argv[0], "brute-force 'steghide extract -sf image.jpg -xf output.txt -p <PASSWORD> -f' resources/passwords.txt\n") print("") sys.exit(0) attacks.brute_force(sys.argv[2], sys.argv[3]) # }}} # {{{ hpf elif sys.argv[1]=="hpf": if len(sys.argv)!=4: print(sys.argv[0], "hpf <input-image> <output-image>\n") print("") sys.exit(0) attacks.high_pass_filter(sys.argv[2], sys.argv[3]) # }}} # {{{ print-diffs elif sys.argv[1]=="print-diffs": if len(sys.argv)!=4: print(sys.argv[0], "print-diffs <cover image> <stego image>\n") print("") sys.exit(0) cover = utils.absolute_path(sys.argv[2]) stego = utils.absolute_path(sys.argv[3]) if not os.path.isfile(cover): print("Cover file not found:", cover) sys.exit(0) if not os.path.isfile(stego): print("Stego file not found:", stego) sys.exit(0) attacks.print_diffs(cover, stego) # }}} # {{{ print-dct-diffs elif sys.argv[1]=="print-dct-diffs": if len(sys.argv)!=4: print(sys.argv[0], "print-dtc-diffs <cover image> <stego image>\n") print("") sys.exit(0) cover = utils.absolute_path(sys.argv[2]) stego = utils.absolute_path(sys.argv[3]) if not os.path.isfile(cover): print("Cover file not found:", cover) sys.exit(0) if not os.path.isfile(stego): print("Stego file not found:", stego) sys.exit(0) attacks.print_dct_diffs(cover, stego) # }}} # {{{ rm-alpha elif sys.argv[1]=="rm-alpha": if len(sys.argv)!=4: print(sys.argv[0], "rm-alpha <input-image> <output-image>\n") print("") sys.exit(0) attacks.remove_alpha_channel(sys.argv[2], sys.argv[3]) # }}} # -- TOOLS -- # {{{ prepare-ml-experiment elif sys.argv[1]=="prep-ml-exp": if len(sys.argv)<6: #print(sys.argv[0], "prep-ml-exp <cover dir> <output dir> <test size> <sim> <payload> [transf]\n") print(sys.argv[0], "prep-ml-exp <cover dir> <output dir> <test size> <sim> <payload>\n") #print(" transf: list of transformations separated by '|' to apply before hidding data.") #print(" - cropNxN: Crop a centered NxN patch") #print(" - resizeNxN: Resize the image to NxN") #print("") print("Example:") #print("", sys.argv[0], " prep-ml-exp cover/ out/ 0.1 hill-sim 0.4 crop512x512|BLUR|SHARP") print("", sys.argv[0], " prep-ml-exp cover/ out/ 0.1 hill-sim 0.4") print("") sys.exit(0) cover_dir = sys.argv[2] output_dir = sys.argv[3] test_size = float(sys.argv[4]) emb_sim = sys.argv[5] payload = float(sys.argv[6]) trn_cover = os.path.join(output_dir, 'trnset', 'cover') trn_stego = os.path.join(output_dir, 'trnset', 'stego') tst_cover = os.path.join(output_dir, 'tstset', 'cover') tst_stego = os.path.join(output_dir, 'tstset', 'stego') tst_cover_to_stego = os.path.join(output_dir, 'tstset', '_cover_to_stego') fn_sim=stegosim.embedding_fn(emb_sim) files = sorted(glob.glob(os.path.join(cover_dir, '*'))) random.seed(0) random.shuffle(files) test_files = files[:int(len(files)*test_size)] train_files = files[int(len(files)*test_size):] print("Using", len(train_files)*2, "files for training and", len(test_files), "for testing.") for d in [trn_cover, trn_stego, tst_cover, tst_stego, tst_cover_to_stego]: try: os.makedirs(d) except: pass for f in train_files: shutil.copy(f, trn_cover) stegosim.embed_message(fn_sim, trn_cover, payload, trn_stego) # to avoid leaks we do not use the same images as cover and stego cover_test_files = test_files[:int(len(test_files)*0.5)] cover_to_stego_test_files = test_files[int(len(test_files)*0.5):] for f in cover_test_files: shutil.copy(f, tst_cover) for f in cover_to_stego_test_files: shutil.copy(f, tst_cover_to_stego) stegosim.embed_message(fn_sim, tst_cover_to_stego, payload, tst_stego) shutil.rmtree(tst_cover_to_stego) # }}} # -- PLOT -- # {{{ plot-histogram elif sys.argv[1]=="plot-histogram": if len(sys.argv)<2: print(sys.argv[0], "plot-histogram <image>\n") print("") sys.exit(0) fn = utils.absolute_path(sys.argv[2]) I = imageio.imread(fn) data = [] if len(I.shape) == 1: data.append(I.flatten()) else: for i in range(I.shape[2]): data.append(I[:,:,i].flatten()) plt.hist(data, range(0, 255), color=["r", "g", "b"]) plt.show() # }}} # {{{ plot-histogram-diff elif sys.argv[1]=="plot-histogram-diff": if len(sys.argv)<4: print(sys.argv[0], "plot-histogram <image> <L|R|U|D>\n") print("") sys.exit(0) fn = utils.absolute_path(sys.argv[2]) direction = sys.argv[3] if direction not in ["L", "R", "U", "D"]: print("Please provide the substract direction: L, R, U or D\n") sys.exit(0) I = imageio.imread(fn) data = [] if len(I.shape) == 1: if direction == "L": D = I[:,1:]-I[:,:-1] if direction == "R": D = I[:,:-1]-I[:,1:] if direction == "U": D = I[:-1,:]-I[1:,:] if direction == "D": D = I[1:,:]-I[:-1,:] data.append(D.flatten()) else: for i in range(I.shape[2]): if direction == "L": D = I[:,1:,i]-I[:,:-1,i] if direction == "R": D = I[:,:-1,i]-I[:,1:,i] if direction == "U": D = I[:-1,:,i]-I[1:,:,i] if direction == "D": D = I[1:,:,i]-I[:-1,:,i] data.append(D.flatten()) plt.hist(data, range(0, 255), color=["r", "g", "b"]) plt.show() # }}} # {{{ plot-dct-histogram elif sys.argv[1]=="plot-dct-histogram": if len(sys.argv)<2: print(sys.argv[0], "plot-dct-histogram <image>\n") print("") sys.exit(0) fn = utils.absolute_path(sys.argv[2]) name, ext = os.path.splitext(fn) if ext.lower() not in [".jpeg", ".jpg"] or not os.path.isfile(fn): print("Please, provide a a JPEG image!\n") sys.exit(0) I = jpeg.JPEG(fn) channels = ["r", "g", "b"] dct_list = [] for i in range(I.components()): dct = I.coeffs(i).flatten() dct_list.append(dct) #counts, bins = np.histogram(dct, range(-5, 5)) #plt.plot(bins[:-1], counts, channels[i]) plt.hist(dct_list, range(-5, 5), rwidth=1, color=["r", "g", "b"]) plt.show() # }}} else: print("Wrong command!")
def main(): attacks_doc="\n" \ " Statistical attacks:\n" \ " - spa: Sample Pairs Analysis.\n" \ " - rs: RS attack.\n" \ " - calibration: Calibration attack to JPEG images." embsim_doc="\n" \ " Embedding simulators:\n" \ " - lsbr-sim: Embedding using LSB replacement simulator.\n" \ " - lsbm-sim: Embedding using LSB matching simulator.\n" \ " - hugo-sim: Embedding using HUGO simulator.\n" \ " - wow-sim: Embedding using WOW simulator.\n" \ " - s-uniward-sim: Embedding using S-UNIWARD simulator.\n" \ " - j-uniward-sim: Embedding using J-UNIWARD simulator.\n" \ " - j-uniward-color-sim: Embedding using J-UNIWARD color simulator.\n" \ " - hill-sim: Embedding using HILL simulator.\n" \ " - ebs-sim: Embedding using EBS simulator.\n" \ " - ebs-color-sim: Embedding using EBS color simulator.\n" \ " - ued-sim: Embedding using UED simulator.\n" \ " - ued-color-sim: Embedding using UED color simulator.\n" \ " - nsf5-sim: Embedding using nsF5 simulator.\n" \ " - nsf5-color-sim: Embedding using nsF5 color simulator." model_doc="\n" \ " Model training:\n" \ " - esvm: Ensemble of Support Vector Machines.\n" \ " - e4s: Ensemble Classifiers for Steganalysis.\n" \ " - srnet: Steganalysis Residual Network." mldetect_doc="\n" \ " ML-based detectors:\n" \ " - esvm-predict: Predict using eSVM.\n" \ " - e4s-predict: Predict using EC.\n" \ " - srnet-predict: Predict using SRNet." feaextract_doc="\n" \ " Feature extractors:\n" \ " - srm: Full Spatial Rich Models.\n" \ " - hill-maxsrm: Selection-Channel-Aware Spatial Rich Models for HILL.\n" \ " - srmq1: Spatial Rich Models with fixed quantization q=1c.\n" \ " - scrmq1: Spatial Color Rich Models with fixed quantization q=1c.\n" \ " - gfr: JPEG steganalysis with 2D Gabor Filters." auto_doc="\n" \ " Unsupervised attacks:\n" \ " - ats: Artificial Training Sets." tools_doc="\n" \ " Tools:\n" \ " - brute-force: Brute force attack using a list of passwords.\n" \ " - hpf: High-pass filter.\n" \ " - print-diffs: Differences between two images.\n" \ " - print-dct-diffs: Differences between the DCT coefficients of two JPEG images.\n" \ " - rm-alpha: Opacity of the alpha channel to 255.\n" \ " - prep-ml-exp: Prepare an experiment for testing ML tools." if len(sys.argv) < 2: print(sys.argv[0], "<command>\n") print("COMMANDS:") print(attacks_doc) print(mldetect_doc) print(feaextract_doc) print(embsim_doc) print(model_doc) print(auto_doc) print(tools_doc) print("\n") sys.exit(0) if False: pass # -- ATTACKS -- # {{{ spa elif sys.argv[1] == "spa": if len(sys.argv) != 3: print(sys.argv[0], "spa <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) threshold = 0.05 I = misc.imread(sys.argv[2]) if len(I.shape) == 2: bitrate = attacks.spa(sys.argv[2], None) if bitrate < threshold: print("No hidden data found") else: print("Hiden data found"), bitrate else: bitrate_R = attacks.spa(sys.argv[2], 0) bitrate_G = attacks.spa(sys.argv[2], 1) bitrate_B = attacks.spa(sys.argv[2], 2) if bitrate_R < threshold and bitrate_G < threshold and bitrate_B < threshold: print("No hidden data found") sys.exit(0) if bitrate_R >= threshold: print("Hiden data found in channel R", bitrate_R) if bitrate_G >= threshold: print("Hiden data found in channel G", bitrate_G) if bitrate_B >= threshold: print("Hiden data found in channel B", bitrate_B) sys.exit(0) # }}} # {{{ rs elif sys.argv[1] == "rs": if len(sys.argv) != 3: print(sys.argv[0], "spa <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) threshold = 0.05 I = misc.imread(sys.argv[2]) if len(I.shape) == 2: bitrate = attacks.rs(sys.argv[2], None) if bitrate < threshold: print("No hidden data found") else: print("Hiden data found", bitrate) else: bitrate_R = attacks.rs(sys.argv[2], 0) bitrate_G = attacks.rs(sys.argv[2], 1) bitrate_B = attacks.rs(sys.argv[2], 2) if bitrate_R < threshold and bitrate_G < threshold and bitrate_B < threshold: print("No hidden data found") sys.exit(0) if bitrate_R >= threshold: print("Hiden data found in channel R", bitrate_R) if bitrate_G >= threshold: print("Hiden data found in channel G", bitrate_G) if bitrate_B >= threshold: print("Hiden data found in channel B", bitrate_B) sys.exit(0) # }}} # {{{ calibration elif sys.argv[1] == "calibration": if len(sys.argv) != 3: print(sys.argv[0], "calibration <image>\n") sys.exit(0) if not utils.is_valid_image(sys.argv[2]): print("Please, provide a valid image") sys.exit(0) fn = utils.absolute_path(sys.argv[2]) attacks.calibration(fn) # }}} # -- ML-BASED DETECTORS -- # {{{ esvm-predict elif sys.argv[1] == "esvm-predict": if len(sys.argv) != 5: print(sys.argv[0], "esvm-predict <model-file> <feature-extractor> <image/dir>") print(feaextract_doc) sys.exit(0) model_file = sys.argv[2] extractor = sys.argv[3] path = utils.absolute_path(sys.argv[4]) files = [] if os.path.isdir(path): for dirpath, _, filenames in os.walk(path): for f in filenames: path = os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files = [path] clf = pickle.load(open(model_file, "r")) for f in files: X = feaext.extractor_fn(extractor)(f) X = X.reshape((1, X.shape[0])) p = clf.predict_proba(X) print(p) if p[0][0] > 0.5: print(os.path.basename(f), "Cover, probability:", p[0][0]) else: print(os.path.basename(f), "Stego, probability:", p[0][1]) # }}} # {{{ e4s-predict elif sys.argv[1] == "e4s-predict": if len(sys.argv) != 5: print( sys.argv[0], "e4s-predict <model-file> <feature-extractor> <image/dir>\n") print("") print(feaextract_doc) print("") sys.exit(0) model_file = sys.argv[2] extractor = sys.argv[3] path = utils.absolute_path(sys.argv[4]) files = [] if os.path.isdir(path): for dirpath, _, filenames in os.walk(path): for f in filenames: path = os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files = [path] clf = models.Ensemble4Stego() clf.load(model_file) for f in files: X = feaext.extractor_fn(extractor)(f) X = X.reshape((1, X.shape[0])) p = clf.predict(X) if p[0] == 0: print(os.path.basename(f), "Cover") else: print(os.path.basename(f), "Stego") # }}} # {{{ srnet-predict elif sys.argv[1] == "srnet-predict": if len(sys.argv) < 4: print(sys.argv[0], "srnet-predict <model dir> <image/dir> [dev]\n") print(" dev: Device: GPU Id or 'CPU' (default='CPU')") print("") sys.exit(0) model_dir = sys.argv[2] path = utils.absolute_path(sys.argv[3]) if len(sys.argv) < 5: dev_id = "CPU" print("'dev' not provided, using:", dev_id) else: dev_id = sys.argv[4] files = [] if os.path.isdir(path): for dirpath, _, filenames in os.walk(path): for f in filenames: path = os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, please provide a valid image: ", f) else: files.append(path) else: files = [path] models.nn_configure_device(dev_id) pred = models.nn_predict(models.SRNet, files, model_dir, batch_size=20) #print(pred) for i in range(len(files)): if pred[i] == 0: print(os.path.basename(files[i]), "Cover") else: print(os.path.basename(files[i]), "Stego") # }}} # -- FEATURE EXTRACTORS -- # {{{ srm elif sys.argv[1] == "srm": if len(sys.argv) != 4: print(sys.argv[0], "srm <image/dir> <output-file>\n") sys.exit(0) image_path = sys.argv[2] ofile = sys.argv[3] feaext.extract_features(feaext.SRM_extract, image_path, ofile) # }}} # {{{ srmq1 elif sys.argv[1] == "srmq1": if len(sys.argv) != 4: print(sys.argv[0], "srmq1 <image/dir> <output-file>\n") sys.exit(0) image_path = sys.argv[2] ofile = sys.argv[3] feaext.extract_features(feaext.SRMQ1_extract, image_path, ofile) # }}} # {{{ scrmq1 elif sys.argv[1] == "scrmq1": if len(sys.argv) != 4: print(sys.argv[0], "scrmq1 <image/dir> <output-file>\n") sys.exit(0) image_path = sys.argv[2] ofile = sys.argv[3] feaext.extract_features(feaext.SCRMQ1_extract, image_path, ofile) # }}} # {{{ gfr elif sys.argv[1] == "gfr": if len(sys.argv) < 4: print(sys.argv[0], "gfr <image/dir> <output-file> [quality] [rotations]\n") sys.exit(0) image_path = sys.argv[2] ofile = sys.argv[3] if len(sys.argv) < 5: quality = "auto" print("JPEG quality not provided, using detection via 'identify'") else: quality = sys.argv[4] if len(sys.argv) < 6: rotations = 32 print("Number of rotations for Gabor kernel no provided, using:", \ rotations) else: rotations = sys.argv[6] params = {"quality": quality, "rotations": rotations} feaext.extract_features(feaext.GFR_extract, image_path, ofile, params) # }}} # {{{ hill-sigma-spam-psrm elif sys.argv[1] == "hill-sigma-spam-psrm": if len(sys.argv) != 4: print(sys.argv[0], "hill-sigma-spam-psrm <image/dir> <output-file>\n") sys.exit(0) image_path = sys.argv[2] ofile = sys.argv[3] feaext.extract_features(feaext.HILL_sigma_spam_PSRM_extract, image_path, ofile) # }}} # {{{ hill-maxsrm elif sys.argv[1] == "hill-maxsrm": if len(sys.argv) != 4: print(sys.argv[0], "hill-maxsrm <image/dir> <output-file>\n") sys.exit(0) image_path = sys.argv[2] ofile = sys.argv[3] feaext.extract_features(feaext.HILL_MAXSRM_extract, image_path, ofile) # }}} # -- EMBEDDING SIMULATORS -- # {{{ lsbr-sim elif sys.argv[1] == "lsbr-sim": if len(sys.argv) != 5: print(sys.argv[0], "lsbr-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.lsbr, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ lsbm-sim elif sys.argv[1] == "lsbm-sim": if len(sys.argv) != 5: print(sys.argv[0], "lsbm-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.lsbm, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ hugo-sim elif sys.argv[1] == "hugo-sim": if len(sys.argv) != 5: print(sys.argv[0], "hugo-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.hugo, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ wow-sim elif sys.argv[1] == "wow-sim": if len(sys.argv) != 5: print(sys.argv[0], "wow-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.wow, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ s-uniward-sim elif sys.argv[1] == "s-uniward-sim": if len(sys.argv) != 5: print(sys.argv[0], "s-uniward-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.s_uniward, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ hill-sim elif sys.argv[1] == "hill-sim": if len(sys.argv) != 5: print(sys.argv[0], "hill-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.hill, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # {{{ j-uniward-sim elif sys.argv[1] == "j-uniward-sim": if len(sys.argv) != 5: print(sys.argv[0], "j-uniward-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.j_uniward, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ j-uniward-color-sim elif sys.argv[1] == "j-uniward-color-sim": if len(sys.argv) != 5: print(sys.argv[0], "j-uniward-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.j_uniward_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ebs-sim elif sys.argv[1] == "ebs-sim": if len(sys.argv) != 5: print(sys.argv[0], "ebs-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ebs, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ebs-color-sim elif sys.argv[1] == "ebs-color-sim": if len(sys.argv) != 5: print(sys.argv[0], "ebs-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ebs_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ued-sim elif sys.argv[1] == "ued-sim": if len(sys.argv) != 5: print(sys.argv[0], "ued-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ued, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ ued-color-sim elif sys.argv[1] == "ued-color-sim": if len(sys.argv) != 5: print(sys.argv[0], "ued-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.ued_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ nsf5-sim elif sys.argv[1] == "nsf5-sim": if len(sys.argv) != 5: print(sys.argv[0], "nsf5-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.nsf5, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ nsf5-color-sim elif sys.argv[1] == "nsf5-color-sim": if len(sys.argv) != 5: print(sys.argv[0], "nsf5-color-sim <image/dir> <payload> <output-dir>\n") sys.exit(0) stegosim.embed_message(stegosim.nsf5_color, sys.argv[2], sys.argv[3], sys.argv[4], embed_fn_saving=True) # }}} # {{{ experimental-sim elif sys.argv[1] == "experimental-sim": if len(sys.argv) != 5: print(sys.argv[0], "experimental-sim <image/dir> <payload> <output-dir>") print( "NOTE: Please, put your EXPERIMENTAL.m file into external/octave\n" ) sys.exit(0) stegosim.embed_message(stegosim.experimental, sys.argv[2], sys.argv[3], sys.argv[4]) # }}} # -- MODEL TRAINING -- # {{{ esvm elif sys.argv[1] == "esvm": if len(sys.argv) != 5: print(sys.argv[0], "esvm <cover-fea> <stego-fea> <model-file>\n") sys.exit(0) from sklearn.model_selection import train_test_split cover_fea = sys.argv[2] stego_fea = sys.argv[3] model_file = utils.absolute_path(sys.argv[4]) X_cover = pandas.read_csv(cover_fea, delimiter=" ").values X_stego = pandas.read_csv(stego_fea, delimiter=" ").values #X_cover=numpy.loadtxt(cover_fea) #X_stego=numpy.loadtxt(stego_fea) X = numpy.vstack((X_cover, X_stego)) y = numpy.hstack(([0] * len(X_cover), [1] * len(X_stego))) X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.10) clf = models.EnsembleSVM() clf.fit(X_train, y_train) val_score = clf.score(X_val, y_val) pickle.dump(clf, open(model_file, "wb")) print("Validation score:", val_score) # }}} # {{{ e4s elif sys.argv[1] == "e4s": if len(sys.argv) != 5: print(sys.argv[0], "e4s <cover-fea> <stego-fea> <model-file>\n") sys.exit(0) from sklearn.model_selection import train_test_split cover_fea = sys.argv[2] stego_fea = sys.argv[3] model_file = utils.absolute_path(sys.argv[4]) X_cover = pandas.read_csv(cover_fea, delimiter=" ").values X_stego = pandas.read_csv(stego_fea, delimiter=" ").values #X_cover=numpy.loadtxt(cover_fea) #X_stego=numpy.loadtxt(stego_fea) X = numpy.vstack((X_cover, X_stego)) y = numpy.hstack(([0] * len(X_cover), [1] * len(X_stego))) X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.10) clf = models.Ensemble4Stego() clf.fit(X_train, y_train) val_score = clf.score(X_val, y_val) clf.save(model_file) print("Validation score:", val_score) # }}} # {{{ srnet elif sys.argv[1] == "srnet": if len(sys.argv) < 5: print( sys.argv[0], "srnet <cover-dir> <stego-dir> <model-name> [dev] [ES] [valsz] [logdir]\n" ) print(" dev: Device: GPU Id or 'CPU' (default='CPU')") print(" ES: early stopping iterations (default=100)") print(" valsz: Size of validation set. (default=0.1%)") print(" logdir: Log directory. (default=log)") print("") sys.exit(0) cover_dir = sys.argv[2] stego_dir = sys.argv[3] model_name = sys.argv[4] if len(sys.argv) < 6: dev_id = "CPU" print("'dev' not provided, using:", dev_id) else: dev_id = sys.argv[5] if len(sys.argv) < 7: early_stopping = 100 print("'ES' not provided, using:", early_stopping) else: early_stopping = int(sys.argv[6]) if len(sys.argv) < 8: val_size = 0.1 print("'valsz' not provided, using:", val_size) else: val_size = int(sys.argv[7]) if len(sys.argv) < 9: log_dir = 'log' print("'logdir' not provided, using:", log_dir) else: log_dir = sys.argv[8] if dev_id == "CPU": print("Running with CPU. It could be very slow!") models.nn_configure_device(dev_id) from sklearn.model_selection import train_test_split cover_files = sorted(glob.glob(os.path.join(cover_dir, '*'))) stego_files = sorted(glob.glob(os.path.join(stego_dir, '*'))) train_cover_files, valid_cover_files, train_stego_files, valid_stego_files = \ train_test_split(cover_files, stego_files, test_size=val_size, random_state=0) print("Using", len(train_cover_files) * 2, "samples for training and", len(valid_cover_files) * 2, "for validation.") output_dir = os.path.join(log_dir, 'output') checkpoint_dir = os.path.join(log_dir, 'checkpoint') for d in [output_dir, checkpoint_dir]: try: os.makedirs(d) except: pass data = (train_cover_files, train_stego_files, valid_cover_files, valid_stego_files) models.nn_fit(models.SRNet, data, model_name, log_path=output_dir, load_checkpoint=model_name, checkpoint_path=checkpoint_dir, batch_size=20, optimizer=models.AdamaxOptimizer(0.001), early_stopping=early_stopping, valid_interval=1000) # }}} # -- AUTOMATED ATTACKS -- # {{{ ats elif sys.argv[1] == "ats": if len(sys.argv) != 6: print(sys.argv[0], "ats <embed-sim> <payload> <fea-extract> <images>\n") print(embsim_doc) print("") print(feaextract_doc) print("") sys.exit(0) emb_sim = sys.argv[2] payload = sys.argv[3] feaextract = sys.argv[4] A_dir = sys.argv[5] fn_sim = stegosim.embedding_fn(emb_sim) fn_feaextract = feaext.extractor_fn(feaextract) import tempfile B_dir = tempfile.mkdtemp() C_dir = tempfile.mkdtemp() stegosim.embed_message(fn_sim, A_dir, payload, B_dir) stegosim.embed_message(fn_sim, B_dir, payload, C_dir) fea_dir = tempfile.mkdtemp() A_fea = os.path.join(fea_dir, "A.fea") C_fea = os.path.join(fea_dir, "C.fea") feaext.extract_features(fn_feaextract, A_dir, A_fea) feaext.extract_features(fn_feaextract, C_dir, C_fea) A = pandas.read_csv(A_fea, delimiter=" ").values C = pandas.read_csv(C_fea, delimiter=" ").values X = numpy.vstack((A, C)) y = numpy.hstack(([0] * len(A), [1] * len(C))) clf = models.Ensemble4Stego() clf.fit(X, y) files = [] for dirpath, _, filenames in os.walk(B_dir): for f in filenames: path = os.path.abspath(os.path.join(dirpath, f)) if not utils.is_valid_image(path): print("Warning, this is not a valid image: ", f) else: files.append(path) for f in files: B = fn_feaextract(f) B = B.reshape((1, B.shape[0])) p = clf.predict(B) if p[0] == 0: print(os.path.basename(f), "Cover") else: print(os.path.basename(f), "Stego") shutil.rmtree(B_dir) shutil.rmtree(C_dir) shutil.rmtree(fea_dir) # }}} # -- NAIVE ATTACKS -- # {{{ brute-force elif sys.argv[1] == "brute-force": if len(sys.argv) != 4: print(sys.argv[0], "brute-force <unhide command> <passw file>\n") print("Example:") print( sys.argv[0], "brute-force 'steghide extract -sf image.jpg -xf output.txt -p <PASSWORD> -f' resources/passwords.txt\n" ) print("") sys.exit(0) attacks.brute_force(sys.argv[2], sys.argv[3]) # }}} # {{{ hpf elif sys.argv[1] == "hpf": if len(sys.argv) != 4: print(sys.argv[0], "hpf <input-image> <output-image>\n") print("") sys.exit(0) attacks.high_pass_filter(sys.argv[2], sys.argv[3]) # }}} # {{{ print-diffs elif sys.argv[1] == "print-diffs": if len(sys.argv) != 4: print(sys.argv[0], "print-diffs <cover image> <stego image>\n") print("") sys.exit(0) cover = utils.absolute_path(sys.argv[2]) stego = utils.absolute_path(sys.argv[3]) if not os.path.isfile(cover): print("Cover file not found:", cover) sys.exit(0) if not os.path.isfile(stego): print("Stego file not found:", stego) sys.exit(0) attacks.print_diffs(cover, stego) # }}} # {{{ print-dct-diffs elif sys.argv[1] == "print-dct-diffs": if len(sys.argv) != 4: print(sys.argv[0], "print-dtc-diffs <cover image> <stego image>\n") print("") sys.exit(0) cover = utils.absolute_path(sys.argv[2]) stego = utils.absolute_path(sys.argv[3]) if not os.path.isfile(cover): print("Cover file not found:", cover) sys.exit(0) if not os.path.isfile(stego): print("Stego file not found:", stego) sys.exit(0) attacks.print_dct_diffs(cover, stego) # }}} # {{{ rm-alpha elif sys.argv[1] == "rm-alpha": if len(sys.argv) != 4: print(sys.argv[0], "rm-alpha <input-image> <output-image>\n") print("") sys.exit(0) attacks.remove_alpha_channel(sys.argv[2], sys.argv[3]) # }}} # -- TOOLS -- # {{{ prepare-ml-experiment elif sys.argv[1] == "prep-ml-exp": if len(sys.argv) < 6: #print(sys.argv[0], "prep-ml-exp <cover dir> <output dir> <test size> <sim> <payload> [transf]\n") print( sys.argv[0], "prep-ml-exp <cover dir> <output dir> <test size> <sim> <payload>\n" ) #print(" transf: list of transformations separated by '|' to apply before hidding data.") #print(" - cropNxN: Crop a centered NxN patch") #print(" - resizeNxN: Resize the image to NxN") #print("") print("Example:") #print("", sys.argv[0], " prep-ml-exp cover/ out/ 0.1 hill-sim 0.4 crop512x512|BLUR|SHARP") print("", sys.argv[0], " prep-ml-exp cover/ out/ 0.1 hill-sim 0.4") print("") sys.exit(0) cover_dir = sys.argv[2] output_dir = sys.argv[3] test_size = float(sys.argv[4]) emb_sim = sys.argv[5] payload = float(sys.argv[6]) trn_cover = os.path.join(output_dir, 'trnset', 'cover') trn_stego = os.path.join(output_dir, 'trnset', 'stego') tst_cover = os.path.join(output_dir, 'tstset', 'cover') tst_stego = os.path.join(output_dir, 'tstset', 'stego') tst_cover_to_stego = os.path.join(output_dir, 'tstset', '_cover_to_stego') fn_sim = stegosim.embedding_fn(emb_sim) files = sorted(glob.glob(os.path.join(cover_dir, '*'))) random.seed(0) random.shuffle(files) test_files = files[:int(len(files) * test_size)] train_files = files[int(len(files) * test_size):] print("Using", len(train_files) * 2, "files for training and", len(test_files), "for testing.") for d in [ trn_cover, trn_stego, tst_cover, tst_stego, tst_cover_to_stego ]: try: os.makedirs(d) except: pass for f in train_files: shutil.copy(f, trn_cover) stegosim.embed_message(fn_sim, trn_cover, payload, trn_stego) # to avoid leaks we do not use the same images as cover and stego cover_test_files = test_files[:int(len(test_files) * 0.5)] cover_to_stego_test_files = test_files[int(len(test_files) * 0.5):] for f in cover_test_files: shutil.copy(f, tst_cover) for f in cover_to_stego_test_files: shutil.copy(f, tst_cover_to_stego) stegosim.embed_message(fn_sim, tst_cover_to_stego, payload, tst_stego) shutil.rmtree(tst_cover_to_stego) # }}} else: print("Wrong command!")