RU2608150C2 - Method of hidden data transfer in video image - Google Patents

Abstract

FIELD: information technology.

SUBSTANCE: invention relates to steganography and is aimed at arrangement of a channel for hidden transfer of additional information in a video image. Proposed is a method of hidden data transfer in a video image as per MPEG-2 standard based on replacing less significant bits of a video image frame with values of a two-dimensional nonlinear code combination containing the hidden information. Formation of the steganographic channel is started with embedded data processing including encoding and modulation with a pseudo-random signal, for which selected are nonlinear 2D Frank-Walsh signals and/or Frank-Krestenson signals. Simultaneously with formation of the stegosignal selected are frames for its embedding considering as suitable all I-frames, as well as B- and P-frames. Stegochannel data are embedded only in those DCT coefficients, which are located in the vicinity of the right diagonal of the DCT coefficients matrix recorded in a JPEG file and added with extra system information and with universal Huffman tables by modulo two addition of the DCT coefficient bits.

EFFECT: technical result is enabling minimization of distortion of the video image, which is being introduced in, while providing stego-resistance of the information transmission system.

1 cl, 7 dwg

Description

The invention relates to the field of steganography, and in particular to the issue of covert transmission of additional information due to the redundancy of video data compressed according to the MPEG-2 standard. Embedding a steganographic channel in a video image before compression according to the MPEG-2 standard is carried out by compiling a list of frames of an elementary video stream suitable for embedding and modifying them with data from a steganographic channel organized on the basis of the Frank-Walsh (F-U), Frank-Crestonson two-dimensional non-linear signal system (FC) [1-4]. This invention is directed to the organization of a channel for covert information transmission due to the redundancy of video data with compression according to the MPEG-2 standard.

A known analogue method of embedding information in a video image proposed in [5-8], based on the addition of a pseudo-random array to low-frequency and high-frequency coefficients of video data compressed according to the MPEG-2 standard. In the process of embedding a stegochannel, only brightness values in I-, B- and P-frames are directly involved.

A similar feature of this method [5-8] with the claimed is the modulation of the data of the stegochannel with a pseudo-random sequence. The difference between the analogue method lies in the search algorithm for embed codewords suitable for embedding with stochannels I _{x, y} (i) and codewords necessary for encoding images of the stegocontainer I _{Wx, y} (i), based on the use of a variable length code table B.14 and B.15 of the MPEG-2 standard and comparing codeword sizes Sz _I and Sz _Iw, respectively. The analogue method provides for the replacement of a codeword intended to encode a stigma, if its size is less than or equal to the length of the codeword, the initial coefficient of DCT. To extract the stegochannel data, the video stream must be fully decoded, and the bits of the year-old must be extracted by calculating the correlation between the stego image and the stego container. The main disadvantage of the analogue, which impedes the achievement of the technical result, is the direct modification of DCT coefficients in the compressed video stream, which leads to the accumulation of shift or errors, as well as low steganographic resistance to the fact of detection of the stegochannel associated with a change in low-frequency DC coefficients.

Closest to the claimed method is the prototype method described in [5, 9-11], based on embedding a steganographic message in the least significant bit. The embedding of secretly transmitted information, consisting of l bits of a certain sequence b _j (j = 0, 1, 2, ..., l-1), into the video data stream compressed according to the MPEG-2 standard is carried out by replacing the least significant bit of the digitized value of the code words by the value of b _j . Modifications are only codewords for which there is at least one other codeword that satisfies the conditions:

- the same length of the zero series;

- the difference between the values of the DCT coefficients is 1;

- the same length of code words.

The embedment process uses only AC current ratios. The procedure for replacing the least significant bits with bits of a steganographic message continues until all bits of this message are embedded. Extraction of the CEH is implemented in a similar way: first, suitable codewords are searched for, from which the low-order bits are taken.

A similar feature of this method [5, 9-11] with the claimed one is the principle of embedding steganographic channel data, based on the replacement of less significant bits of the video image.

The known prototype method has a significant drawback consisting in low steganographic resistance to detecting the fact of transmission, extraction, removal and modification of embedded information.

The inventive method of embedding a steganographic channel in a video image solves the problem of the hidden transmission of information while providing steganographic stability, estimated by the time of identification of the structure of the signal-base of the stego channel.

The required technical result of the implementation of the proposed method of embedding a steganographic channel in a video image before compression according to the MPEG-2 standard should be considered as minimizing distortions of the video image into which it is implemented, while ensuring sufficient steg resistance of the information transmission system estimated by the time it takes to identify the steganographic channel in the video data structure by third parties [1- four].

An essential feature characterizing the claimed method and providing the specified technical result is the combination of the following sequentially performed actions shown in FIG. 1, FIG. 2.

1. The formation of the steganographic channel is carried out on the basis of two-dimensional non-linear Frank-Walsh and (or) Frank-Crestonson signals, which form an orthonormal basis in the case of a dyadic shift and in the case of an n-shift paired with fast Walsh and Vilenkin-Crestonson transformations, respectively [1- four]. At the initial stage, the information sequence is transformed into the binary form a _j , a _j ∈ {-1,1}, j∈N. Then, the obtained discrete signal is “expanded” s times by multiplying the vector of the original message by a unit matrix of the corresponding dimension, resulting in the sequence b _i = a _j , js≤i <(j + 1) s, i∈N. The purpose of the extension is to introduce redundancy, one bit of the stego channel is embedded in s pixels of the image spectrum. Further, this sequence can be multiplied by adaptively selectable factors, after which it is modulated by the pseudo-noise signal of Frank-Walsh (FC) or Frank-Crestonson (FC) p _i , p _i ∈ {-1,1}, i∈N , the purpose of which is to expand the spectrum of the stegokanal. The choice of a signal system for forming a covert channel in the video data structure is justified by the following features of the F-U (F-K) signals [12-16]:

- ease of implementation of the encoder of the transmitter, namely the possibility of forming a system of signals from one;

- the ability to restore the signal on the receiving side with losses no worse than every 30th symbol in the period;

- the existence and uniqueness of the filter for suppressing side lobes for these signals;

- a uniform energy spectrum and a wide signal base, ensuring the secrecy of embedding data in a video image.

After modulation with the F-U (F-K) signals, the stegochannel data is located on the right diagonal of the F-U (F-K) signal matrix.

2. Demultiplexing of the original transport video stream generated in accordance with the MPEG-2 standard is carried out according to the algorithm shown in FIG. 3. Further operations are carried out on the selected elementary video stream.

3. The selection of frames suitable for modification by the signal of the stego channel. To do this, initially form a list of frames suitable for embedding. All I-frames are considered a priori suitable for embedding. As a criterion for assessing the possibility of embedding in B- or P-frames, the ratio of the number of non-zero DCT coefficients in the frame to the number of non-zero DCT coefficients in the previous I-frame (the first frame of the group) is selected. Marking this ratio as R, you get a selection rule for B- or P-frames

. Алгоритм анализа исходного видеофайла представлен на фиг. 4.where N _{B, P} is the number of DCT coefficients γ _{i, j} ≠ 0, i = 0 ... 6, j = 0 ... 64 in the current B or P frame, N _I is the number of DCT coefficients γ _{i, j} ≠ 0, i = 0 ... 6, j = 0 ... 64 in the previous I-frame. A B or P frame is considered suitable for embedding if

. The algorithm for analyzing the source video file is shown in FIG. four.

4. At the stage of forming frames in JPEG format, the quantized values of DCT coefficients of the corresponding frames of the video image are downloaded in the form of graphic JPEG files for further processing by the steganograph. At the same time, JPEG files are formed in the buffer, in which the DCT values of the corresponding frame of the elementary video stream are recorded with the mandatory formation of universal Huffman tables. When the MPEG video stream is formed in B- and P-frames, some of the macroblocks containing only zero values are not encoded; instead, a number that determines the number of skipped macroblocks is placed in the stream. In the process of generating JPEG files, zero values are recorded at the place of the missed macroblocks, which is necessary to ensure the same frame size regardless of the number of missed macroblocks and additionally ensures the secrecy of data embedding.

To ensure uniform distribution of hidden information across the container, as well as to exclude changes in the structure of the transport video stream and out of sync between video and audio data associated with a change in the initial boundaries of the PES packets, data that is not fit in the first video packet is forced into the second, then to the third and so on, without affecting the audio and system information. (Fig. 5). Due to the fact that the sizes of PES packets do not change, the secrecy of the bookmark is increased.

After processing the JPEG file with a steganograph, the correct Huffman table is formed taking into account the distribution of coefficient values, as a result of which the compression ratio increases and, accordingly, the size of the JPEG file decreases.

5. Embedding the stegochannel data is carried out by modulo adding two formed stego sequences based on the F-U (F-K) signals and bit words of the DCT coefficients of the video stream uploaded to JPEG files. At the same time, the coefficients located in the vicinity of the right diagonal of the matrix of DCT coefficients are chosen for implementation, since embedding information in the coefficients to the right of the diagonal will lead to their loss, and to the left - to significant image distortions. Modifications are made to bits occupying the second to fourth extreme right positions of an eight-bit word (Fig. 6). The choice of embedding positions is justified by the following facts: a change in the least significant bit is a unmasking sign, and the probability of data loss embedded in the least significant bit is 0.9; modification of bits located at 5-8 positions on the right leads to significant distortion of the video image. The stegosignal modulated by the information message, added to the image spectrum, has the form v _i = v _i + a _i b _i p _i , i∈N.

6. When extracting hidden information, use the same principles as when embedding it, and start by loading the list of frames prepared at the analysis stage. The original video stream is demultiplexed and a search is made for the frames indicated in the list in the obtained elementary video stream. At the same time, an output (resulting) stream is also formed. Before finding the desired frame, data from the buffer of the source stream is placed directly in the output stream in accordance with the movement of the marker during demultiplexing.

When finding the desired frame, perform the following actions.

1. Download the JPEG file corresponding to this frame. After recording the header and a number of service fields, recording in the resulting stream is stopped.

2. Decode the next macroblock in the original video stream. If this is the first macroblock in the slice, then the service field from the source stream is placed in the output stream, since this value cannot be calculated later, then the DCT coefficients corresponding to this macroblock are downloaded from the JPEG file and subjected to entropy encoding (Fig. 7) . The coding results are placed in the output stream; the process is repeated for all macroblocks of the frame. During decoding, the data of the next PES packet located in the decoder buffer may end, and the task of finding and loading a new PES packet with video data into the buffer will arise. If the current PES packet in the output stream is not completely filled with video data, the system and audio information coming from the source stream when searching for the PES packet with video data is not written to the output stream, but is placed in the buffer. The contents of the buffer are placed in the output stream after the next PES packet is completely filled with video data. Similarly, if the amount of video data coming from the entropy encoder exceeds the size of the current PES packet, the video data is buffered, and PES packets with system and audio information are written to the output stream until the demultiplexer detects the next PES packet with video data. The contents of the buffer are dumped into the output stream, video data is placed there, and the buffer is used to store system information.

3. The extraction of the stegochannel data is carried out using a side-lobe suppression filter, which makes it possible to decide on the transmitted signal even in the case of a signal-to-noise ratio of less than unity.

To retrieve stegoknal data on the receiving side, the recipient must have the following key information:

- the type of signals used in the formation of the stegochannel;

- a list of frames selected for embedding;

- embedment rules, selection of DCT coefficients, selection of bit positions subjected to modification);

- cipherdata.

The delivery of key information to the correspondent is entrusted to the user. With insufficient container volume, the bookmark is previously divided into several parts, interleaved. In this case, the correspondent is informed of the assembly procedure of the original message. One way to solve this problem is to place this information on a bookmark basis in one of the frames, the number of which is transmitted to the correspondent along with the key.

Due to the pseudo-noise nature of the Ф-У, Ф-К pi signals, the stegosignal is also pseudo-noise and it is difficult to detect it both by visual and statistical methods. The general requirement for sequences is orthogonality. The CEH demodulation is performed by a correlation receiver represented by a two-dimensional side-lobe suppression filter, the existence and uniqueness of which was proved in [1, 3, 16]. In the FPL, the pixels of the stego image are multiplied by the same SHPS as in the encoder, and then the result is summed over all the samples containing the data of the stegochannel:

, т.е. фильтрация не повлияла на данные стегоканала.Here, the sums ∑ ₁ and ∑ ₂ reflect the contribution to the correlation sum of the filtered image and the filtered stegochannel data, respectively. Suppose that ∑ ₁ is zero, i.e. the image is completely filtered by the high-pass filter, and

, i.e. filtering did not affect stegochannel data.

, где

- дисперсия псевдошумовой последовательности. Знак корреляционной суммы есть внедренный бит информации a _i, так что информацию извлекают без потерь. Этот сигнал умножают почленно на псевдошумовой сигнал. Математическое ожидание произведения равно нулю, а дисперсия

. Это произведение суммируется L раз. Если L достаточно большое, то в соответствии с центральной предельной теоремой плотность распределения вероятностей суммы стремится к нормальному распределению с нулевым матожиданием и дисперсией

. Ошибка в определении бита возникнет, если информационный бит был +1, а

, или если бит был равен -1, а

. Так как сумму описывают нормальным распределением, вероятность ошибки на бит будет равна:Then the correlation sum is

where

- variance of the pseudo-noise sequence. The sign of the correlation sum is an embedded bit of information a _i , so that information is extracted without loss. This signal is multiplied term by noise. The mathematical expectation of the product is zero, and the variance

. This product is summed up L times. If L is large enough, then, in accordance with the central limit theorem, the probability distribution density of the sum tends to a normal distribution with zero expectation and dispersion

. An error in determining the bit will occur if the information bit was +1, and

, or if the bit was -1, and

. Since the sum is described by the normal distribution, the probability of error per bit will be equal to:

The error probability formula shows that in order to reduce it, it is necessary to increase the coefficient of expansion L, the dispersion of the NPS σ _p, or the average amplitude α _i .

The Frack-Walsh (Frank-Crestonson) signal systems are resistant to time thinning, which ensures their use for embedding into video data during compression, they also have minimal energy due to their delta correlation and introduce minimal distortion to the image, which ensures the achievement of a technical result, namely, that the fact of transmitting information to the video data structure will be revealed in at least tens of hours.

Thus, the essential features that distinguish the claimed method from the prototype are:

1) pre-processing of embedded data, providing for encryption and reduction to a noise-like appearance as a result of modulation by signals Ф-У (Ф-к), in contrast to the prototype, where only data is converted to binary;

2) the choice of frames for embedding stegochannel data, which considers all I-frames suitable for embedding the stegochannel, as well as B- and P-frames, the ratio of the number of non-zero DCT coefficients to the number of non-zero DCT coefficients of the previous I-frame, in which there are more than ¼, in contrast from the prototype, where embedding is carried out only in I-frames;

3) the choice of DCT coefficients for embedding, which consists in changing the coefficients located in the vicinity of the right diagonal, the DCT coefficient matrix, in contrast to the prototype, where only codewords for which there is at least one other codeword with equal length of the zero series are subject to change; with the same length of code words, with a difference between the values of the DCT coefficients per unit;

4) the choice of bit positions, involving the change of bits occupying the second to fourth position on the right, in contrast to the prototype, where only the least significant bit is changed;

5) the rule for embedding stegochannel data, implying modulo addition of two sequences of the stegosignal and a sequence of DCT coefficients of the image, in contrast to the prototype, where embedding is carried out by substitution;

6) the rule for the distribution of stegochannel data on PES packets of video data without changing the initial boundaries, which supposes the displacement of non-placed information in the next PES packet, in contrast to the prototype, where data distribution is not considered;

7) a method for extracting stegochannel data, which involves the use of a side lobe suppression filter, in contrast to the prototype, where the extraction is performed by extracting the values of the least significant bits of the appropriate codewords.

To clarify the essential features of the proposed method are shown figures:

Figure 1 - Block diagram of the algorithm for embedding data in a video stream;

Figure 2 - Scheme of the algorithm of the steganograph program;

Figure 3 - Block diagram of the algorithm for demultiplexing the source stream;

Figure 4 - A block diagram of an algorithm for analyzing the basic information of a video file;

Figure 5 - Redistribution of steganographic channel data between PES packets;

Figure 6 - Modification of bits of the DCT coefficients of the image when embedding the data of the stegochannel;

Figure 7 - Diagram of the algorithm for reading data from a JPEG file.

The possibility of carrying out the invention is confirmed by the known characteristics and functionality of digital signal processors for the implementation of operations performed in the process of embedding and processing data in accordance with the claimed method [17-20].

Literature

1. Discrete harmonic analysis and its applications to the synthesis of optimal signals: monograph / K.Yu. Tsvetkov, V.M. Korovin. - SPb .: VKA them. A.F. Mozhaysky, 2008 .-- 108 p.

2. Tsvetkov K.Yu. The theory of optimal systems of complex discrete signals and its applications. - SPb .: VKA, 2005 .-- 160 s.

3. Tsvetkov K.Yu., Malozemov VN On the optimal signal-filter pair // Problems of information transfer. - 2003. - T. 1. - Vol. 2. - S. 50-62.

4. Tsvetkov K.Yu., Malozemov V.N., Masharsky S.M. Frank signal and its generalization // Problems of information transfer. - 2001. - T. 37. - Issue. 2. - S. 18-26.

5. Gribunin V.G., Okov I.N., Turintsev I.V. Digital steganography. - M .: Solon-Press, 2002 .-- 272 p.

6. Marvel L. Image Steganography for hidden communication. PhD Thesis. Univ. of Delaware, 1999.115p.

7. Cox J., Miller M., McKellips A. Watermarking as communications with side information // Proceedings of the IEEE. 1999. Vol. 87. No. 7. P. 1127-1141.

8. Schneier B. Applied Cryptography: Protocols, Algorithms, and Source Code in C, 2nd ed. New York // John Wiley and Sons, 1996.

9. Anderson R., editor. // Proc. Int. Workshop on Information Hiding: Lecture Notes in Computer Science. Springer-Verlag, Cambridge. 1996.

10. Ramkumar M. Data Hiding in Multimedia. PhD Thesis. New Jersey Institute of Technology, 1999. 72 p.

11. Simmons G. The History of Subliminal Channels // IEEE Journal on Selected Areas of Communications. 1998. Vol. 16, No. 4. P. 452-461.

12. Abazina E.S. Algorithms for the implementation of two-dimensional nonlinear code sequences in the structure of compressed video data // Questions of radio electronics in the field of television technology, 2013, No. 1, p. 85-94.

13. Abazina E.S. Algorithms for processing broadband digital watermarks when organizing a steganographic channel in the structure of video data // Transactions of the Military Academy of Communications, Issue 79. - SPb: YOU: 2013. - P. 9-14.

14. Korovin V.M. Method and algorithms for embedding broadband digital watermarks in compressed images // Sat. reports of the international scientific-practical conference "Features of the development of the Russian space industry and prospects for its further integration into the system of international economic relations." October 2-3, 2007, St. Petersburg. St. Petersburg: BSTU, 2007 .-- S. 175-178.

15. Korovin V.M., Tsvetkov K.Yu. Synthesis of optimal two-dimensional signals and filters for suppressing the side lobes of the correlation functions of complex discrete signals in the Vilenkin – Chrestenson basis // Aerospace Instrumentation. - 2008. - No. 12. - S. 19-23.

16. Korovin V.M. Optimal two-dimensional filters for suppressing the side lobes of the correlation functions of complex discrete signals in the Vilenkin – Chrestenson basis // Collection of Algorithms and Programs of Typical Problems / Ed. I.A. Kudryashova. - SPb .: VKA them. A.F. Mozhaysky, 2008. - Vol. 27.- S. 286-294.

17. Lokshin B.A. Digital television: from studio to viewer. - M .: Company CYRUS SYSTEMS, 2001. - 446 S.

18. Goldenberg L.N., Matyushkin B.D., Polyak M.N. Digital signal processing. Directory. - M .: Radio and communications, 1990.

19. Ptachek M. Digital television. Theory and technology. - M .: Radio and communications, 1990.

20. Cantor L.Ya. On digital television broadcasting // Telesputnik, No. 11, 2000. - P. 14.

Claims (1)

A method of covert data transmission in a video image according to the MPEG-2 standard, based on changing less significant bits of a video image frame with values of a two-dimensional nonlinear code combination, which carries secretly transmitted information, characterized in that the formation of the steganographic channel begins with the processing of embedded data, including encryption and modulation a pseudo-random signal, in which two-dimensional non-linear Frank-Walsh and / or Frank-Chrestenson signals are selected, simultaneously with the processing of embedded OF DATA frames are selected for their insertion, assuming suitable all I-frames and B- frames and P-ratio of the number of nonzero coefficients of discrete cosine transform (DCT) to the number of non-zero DCT coefficients preceding I-frame in which the longer

; the stegochannel data is embedded only in those DCT coefficients that are located in the vicinity of the right diagonal of the matrix of DCT coefficients recorded in a JPEG file and supplemented by system information and universal Huffman tables by modulo adding two bits of DCT coefficients, occupying the second to fourth positions on the right , with stego bits; in this case, the embedded data is distributed among the PES packets of video data without changing the initial boundaries in the next packet without affecting the audio and system information; the extraction of hidden information begins with loading the list of frames prepared at the stage of analysis, demultiplexing the original video stream and searching for frames indicated in the list; at the same time, an output stream is formed into which, prior to finding the desired frame, data from the buffer of the source stream is placed; when the required frame is found, the corresponding JPEG file is downloaded, and recording in the output stream is stopped, then the DCT coefficients corresponding to this macroblock are downloaded from the JPEG file and subjected to entropy encoding, the encoding results are placed in the output stream, as well as in the side-lane suppression filter for extracting the data of the stegochannel with their subsequent decryption; the process is repeated for all macroblocks of the frame.

Images