基于辨识性统计特征的PQ隐密图像识别算法

doi:10.11959/j.issn.1000-436x.2015068

Abstract

Abstract:

A PQ (perturbed quantization) stego images recognition algorithm is proposed based on identifiable statistical feature. According to the specific changing ways of PQ steganography to image data, the proposed algorithm extracts the identifiable statistical feature that can distinguish PQ stego images from other types of stego images. Then, the SVM (support vector machines) classifier is trained to recognize PQ stego images. Experimental results show that, the proposed algorithm can reliably recognize PQ stego images from multi-class stego images generated by five types of well-known JPEG steganography (PQ、F5ns、F5、MB1 and MOD). Even though the stego images generated by F5、nsF5、MB1 and MOD are not used for training classifier, the proposed algorithm can still effectively recognize PQ stego images.

Key words: steganalysis, stego image recognition, PQ steganography, identifiable statistical feature

Ji-cang LU,Fen-lin LIU,Xiang-yang LUO,Yi ZHANG. Recognition of PQ stego images based on identifiable statistical feature[J]. Journal on Communications, 2015, 36(3): 139-148.

Figures/Tables 9

References 23

[1]	沈昌祥，张焕国，冯登国等．信息安全综述[J]．中国科学E辑：信息科学， 2007,37(2):129-150. SHEN C X , ZHANG H G , FENG D G , et al. Survey of information security[J]. Science in China Series: E, 2007,37(2):129-150.
[2]	王朔中，张新鹏，张卫明．以数字图像为载体的隐写分析研究进展[J]．计算机学报， 2009,32(7):1247-1263. WANG S Z , ZHANG X P , ZHANG W M . Recent advances in im-age-based steganalysis research[J]. Chinese Journal of Computers, 2009,32(7):1247-1263.
[3]	KER A D , BAS P , B?HME R , et al. Moving steganography and stega-nalysis from the laboratory into the real world[A]. Proceedings of the 1st ACM Workshop on Information Hiding and Multimedia Security[C]. Montpellier, France, 2013.45-58.
[4]	PEVNY T , FRIDRICH J . Towards multi-class blind steganalyzer for JPEG images[A]. Proceedings of the 4th International Workshop on Digital Watermarking[C]. Siena, Italy, 2005.39-53.
[5]	PEVNY T , FRIDRICH J . Multiclass detector of current steg-anographic methods for JPEG format[J]. IEEE Transactions on Information Forensics and Security, 2008,3(4):635-650.
[6]	WANG P , LIU F L , WANG G D , et al. Multi-class steganalysis for JPEG stego algorithms[A]. Proceedings of the 15th IEEE International Conference on Image Processing[C]. San Diego, CA, 2008.2076-2079.
[7]	DONG J , WANG W , TAN T N . Multi-class blind steganalysis based on image run-length analysis[A]. Proceedings of the 8th International Workshop on Digital Watermarking[C]. Guildford, UK, 2009.199-210.
[8]	LUO X Y , WANG D S , WANG P , et al. A review on blind detection for image steganography[J]. Signal Processing, 2008,88(9):2138-2157.
[9]	KODOVSKY J , FRIDRICH J . Rich models for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2012,7(3):868-882.
[10]	罗向阳，刘粉林，杨春芳等．基于最优小波包分解的图像隐写通用检测[J]．中国科学：信息科学， 2010,40(2):327-339. LUO X Y , LIU F L , YANG C F , et al. Universal steganalysis based on best wavelet packet decomposition[J]. Science in China Series F: In-formation Sciences, 2010,40(2):327-339.
[11]	FRIDRICH J , GOLJAN M , SOUKAL D . Perturbed quantization steganography with wet paper codes[A]. Proceedings of the 2004 ACM Workshop on Multimedia and Security[C]. Magdeburg, Germany, 2004.4-15.
[12]	GUL G , KURUGOLLU F . SVD-based universal spatial domain image steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2010,5(2):349-353.
[13]	PEVNY T , BAS P , FRIDRICH J . Steganalysis by subtractive pixel adjacency matrix[J] IEEE Transactions on Information Forensics and Security, 2010,5(2):215-224.
[14]	FRIDRICH J , PEVNY T , KODOVSKY J . Statistically undetectable JPEG steganography: dead ends challenges, and opportunities[A]. Proceedings of the 9th ACM Workshop on Multimedia and Security[C]. 2007.3-14.
[15]	ZHANG Y , LIU F L , LUO X Y , et al. Steganalysis of PQ based on the histogram difference between coefficient adjacent positions[A]. Proceedings of the 4th International Conference on Multimedia Information Networking and Security[C]. Nanjing, China, 2012.955-958.
[16]	WESTFELD A . F5—A steganographic algorithm—high capacity despite better steganalysis[A]. Proceedings of the 4th International Workshop on Information Hiding[C]. Pittsburgh, PA, USA, 2001.289-302.
[17]	SALLEE P . Model-based steganography[A]. Proceedings of the 2nd International Workshop on Digital Watermarking[C]. Seoul, Korea, 2003.154-167.
[18]	FILLER T , FRIDRICH J . Design of adaptive steganographic schemes for digital images[A]. Proceedings of SPIE, Electronic Imaging, Security and Forensics of Multimedia XIII[C]. 2011.1-14.
[19]	LUO X Y , LIU F L , LIAN S G , et al. On the typical statistic features for image blind steganalysis[J]. IEEE Journal on Selected Areas in Communications, 2011,29(7):1404-1422.
[20]	陈铭，张茹，刘凡凡等．基于区域相关性的 LSB 匹配隐写分析[J]．通信学报， 2010,31(3):1-11. CHEN M , ZHANG R , LIU F F , et al Steganalysis of LSB matching based on regional correlation[J]. Journal on Communications, 2010,31(3):1-11.
[21]	MALIK H , SUBBALAKSHMI K P , CHANDRAMOULI R . Non-parametric steganalysis of QIM steganography using approximate entropy[J]. IEEE Transactions on Information Forensics and Security, 2012,7(2):418-431.
[22]	SALLEE P . Model-based methods for steganography and steganalysis[J]. International Journal of Image Graphics, 2005,5(1):167-190.
[23]	CHANG C C , LIN C J . LIBSVM: a library for support vector machines[J]. ACM Transactions on Intelligent Systems and Technology, 2011,2(3):1-27.

Metrics

Recommended 0

No Suggested Reading articles found!

图像类别	信息嵌入比率/bpnc	图像数量
载体图像	0	5 000
PQ隐密图像	0.05、0.10、0.15、0.20	5 000×4 = 20 000
F5隐密图像	0.05、0.10、0.15、0.20	5 000×4 = 20 000
nsF5隐密图像	0.05、0.10、0.15、0.20	5 000×4 = 20 000
MB1隐密图像	0.05、0.10、0.15、0.20	5 000×4 = 20 000
MOD隐密图像	0.05、0.10、0.15、0.20	5 000×4 = 20 000

嵌入率/bpnc	PQ vs F5	PQ vs nsF5	PQ vs MB1	PQ vs MOD
0.05	92.47	91.65	92.27	91.87
0.10	97.67	96.93	97.65	97.15
0.15	98.27	97.23	98.51	97.66
0.20	98.55	97.22	98.81	97.94

嵌入率/bpnc	F5 vs nsF5	F5 vs MB1	F5 vs MOD	nsF5 vs MB1	nsF5 vs MOD	MB1 vs MOD
0.05	57.57	63.18	59.05	63.08	56.67	63.13
0.10	68.47	75.46	70.26	77.24	65.58	77.07
0.15	76.11	84.40	78.14	85.40	71.58	85.84
0.20	78.85	90.20	83.47	87.84	74.98	89.39

嵌入率/bpnc	检测算法	PQ	非PQ = (F5 + nsF5 + MB1 + MOD) / 4
	文献[5]算法	90.77	20.87 = (11.31 + 25.01 + 23.31 + 23.85) / 4
0.05	文献[9]算法	99.21	10.17 = (0.55 + 14.29 + 12.51 + 13.31) / 4
	本算法	90.03	6.59 = (5.09 + 7.44 + 6.68 + 7.16) / 4
	文献[5]算法	95.34	18.42 = (4.87 + 25.15 + 23.14 + 20.54) / 4
0.10	文献[9]算法	99.73	9.54 = (0.25 + 13.91 + 11.74 + 12.24) / 4
	本算法	98.52	5.52 = (3.19 + 7.18 + 5.64 + 6.08) / 4
	文献[5]算法	95.46	17.03 = (2.14 + 24.46 + 23.85 + 17.66) / 4
0.15	文献[9]算法	99.73	9.12 = (0.16 + 13.71 + 11.50 + 11.12) / 4
	本算法	98.52	5.52 = (3.19 + 7.18 + 5.64 + 6.08) / 4
	文献[5]算法	95.84	16.54 = (0.96 + 24.06 + 25.97 +15.58) / 4
0.20	文献[9]算法	99.78	8.95 = (0.10 + 13.50 + 11.93 + 10.25) / 4
	本算法	98.68	4.26 = (1.58 + 6.26 + 4.77 + 4.43) / 4

已知隐写算法		平均正确率			平均时间耗费			ROC曲线下面积
已知隐写算法	文献[5]算法	文献[9]算法	本算法	文献[5]算法	文献[9]算法	本算法	文献[5]算法	文献[9]算法	本算法
F5	84.28	92.37	95.08	38:42	31:23	09:49	0.962 2	0.988 5	0.991 6
nsF5	84.50	97.92	96.06	58:38	37:16	11:33	0.944 7	0.996 0	0.991 7
MB1	62.47	84.19	94.46	18:04	33:50	09:59	0.845 9	0.939 2	0.990 1
MOD	65.44	82.43	95.54	29:15	33:39	09:34	0.797 9	0.931 4	0.991 8

Recognition of PQ stego images based on identifiable statistical feature

RichHTML

PDF

Knowledge

Cited

Abstract

Cite this article

share this article

Figures/Tables 9

References 23

Related Articles 15

Metrics

Recommended 0

[1]	Chao XIA, Yaqi LIU, Qingxiao GUAN, Xin JIN, Yanshuo ZHANG, Shengwei XU. Steganalysis of JPEG images using non-linear residuals [J]. Journal on Communications, 2023, 44(1): 142-152.
[2]	Xiaodan WANG, Jingtai LI, Yafei SONG. DDAC: a feature extraction method for model of image steganalysis based on convolutional neural network [J]. Journal on Communications, 2022, 43(5): 68-81.
[3]	Xi SUN,Wei-ming ZHANG,Neng-hai YU,Yao WEI. Steganography based on parameters’ disturbance of spatial image transform [J]. Journal on Communications, 2017, 38(10): 166-174.
[4]	Tao HAN,Xi CHEN,Yue-fei ZHU. Subtractive pixel adjacency matrix based features for steganalysis of spatial color images [J]. Journal on Communications, 2016, 37(2): 158-165.
[5]	Pei-tao YANG,Wei-ming ZHANG,Neng-hai YU. Reducing false positives of steganalysis via classification of image-acquiring sources [J]. Journal on Communications, 2016, 37(12): 165-170.
[6]	Ting DA,Zhi-tang LI. Video steganalysis scheme based on weighted undirected graph [J]. Journal on Communications, 2014, 35(Z1): 27-30.
[7]	. Video steganalysis scheme based on weighted undirected graph [J]. Journal on Communications, 2014, 35(Z1): 6-30.
[8]	Jia-yong CHEN,Jiu-fen LIU,Yue-fei ZHU,Wei-ming ZHANG. Cryptographic secrecy of steganographic matrix encoding [J]. Journal on Communications, 2012, 33(6): 174-179.
[9]	Hai-ying GAO. Steganalysis of quantization index modulation steganography based on the nonuniform DCT [J]. Journal on Communications, 2011, 32(11A): 209-212.
[10]	Ke QI,Da-fang ZHANG,Dong-qing XIE. Reliable steganalysis of color images based on color gradient sequence [J]. Journal on Communications, 2011, 32(1): 27-36.
[11]	Yong-kuan ZHAO,Xiao-xia CAI,Hong CHEN. Fuzzy comprehensive evaluation for image blind steganalysis algorithms [J]. Journal on Communications, 2010, 31(9A): 259-263.
[12]	Jia-yong CHEN,Yue-fei ZHU,Wei-ming ZHANG,Jiu-fen LIU. Chosen-key extracting attack to random LSB steganography [J]. Journal on Communications, 2010, 31(5): 73-80.
[13]	Ming CHEN,Ru ZHANG,Fan-fan LIU,Jia WU,Xin-xin NIU. Steganalysis of LSB matching based on regional correlation [J]. Journal on Communications, 2010, 31(3): 1-11.
[14]	Gang LUO,Xing-ming SUN. Steganalysis for stegotext based on text redundancy [J]. Journal on Communications, 2009, 30(6): 19-25.
[15]	Bin LIU,Fen-lin LIU,Chun-fang YANG. Video steganalysis scheme based on inter-frame collusion [J]. Journal on Communications, 2009, 30(4): 41-49.

嵌入率/bpnc	检测算法	PQ	非PQ = (F5 + nsF5 + MB1 + MOD) / 4
	文献[5]算法	95.31	18.87 = (29.70 + 11.21 + 28.98 + 5.59) / 4
0.05	文献[9]算法	98.66	6.88 = (13.62 + 3.21 + 8.93 + 1.75) / 4
	本算法	93.09	8.96 = (8.29 + 9.02 + 9.66 + 8.85) / 4
	文献[5]算法	98.68	45.08 = (81.91 + 26.09 + 66.25 + 6.05) / 4
0.10	文献[9]算法	99.67	25.98 = (62.70 + 7.27 + 31.98 + 1.95) / 4
	本算法	99.19	9.33 = (8.09 + 9.30 + 11.13 + 8.78) / 4
	文献[5]算法	98.87	60.88 = (97.65 + 50.73 + 88.36 + 6.78) / 4
0.15	文献[9]算法	99.60	43.90 = (94.14 + 18.96 + 60.16 + 2.35) / 4
	本算法	99.29	9.99 = (8.42 + 9.66 + 13.01 + 8.88) / 4
	文献[5]算法	99.06	69.55 = (99.57 + 74.52 + 95.89 + 8.20) / 4
0.20	文献[9]算法	99.71	55.34 = (99.44 + 41.17 + 77.69 + 3.05) / 4
	本算法	99.26	10.83 = (8.95 + 10.04 + 15.55 + 8.79) / 4

检测算法	平均正确率	平均时间耗费	ROC曲线下面积
文献[5]算法	60.72	14:12	0.781 1
文献[9]算法	73.46	13:12	0.910 9
本算法	91.72	04:34	0.988 7