RU2618379C1 - Method of steganographic additional information implementation to samples of digital sound signals - Google Patents

Abstract

FIELD: electricity.

SUBSTANCE: in this method the encrypted concealed information in a binary form is bit-implemented into the samples by replacing the sample bit with the concealed bit of the additional information, samples are used only with a large absolute digital code value, implementing is produced not only in the least significant bits of samples, but in the most significant bits used for the additional information implemention into the sound channels, the number of the sample bits, the interval between the samples used for implemention, and the sequence of the additional information bit implemention into the samples are selected according to the secret pseudo-random key and the psychophysical characteristics of human hearing.

EFFECT: increasing the cryptographic strength of the implemented additional information.

8 dwg, 1 tbl

Description

Analysis of the current status of the issue

The invention relates to the field of telecommunications and is intended for the secret transmission (or storage) of classified information and can be used to protect copyrights (embedding watermarks), secretly transmit passwords, keys, etc.

As is known [7], it is possible to embed (secretly place) information in any digital file that has redundancy.

A known method of embedding watermarks in multimedia files [8]. The method consists in the fact that quantization noise that occurs during the analog-to-digital conversion is removed from the original sound file. Instead of quantization noises, a signal is introduced at the same level as quantization noises, but containing multi-bit labels, with the help of which the protected commercial product is identified (identified). The method is intended to protect copyrights (for example, composers, pop singers, record companies, film studios, television channels, photographers, artists).

The method has good resistance to distortion, conversion from digital to analog and vice versa, fragmentation of a musical work or image, change in volume level, scaling, variation in contrast and brightness, color balance, sampling frequency, etc.

The patent specification details the methodology for proving the unauthorized use of photographs and films.

The performance of this method of covert information transfer in multimedia environments is low, since the introduction is carried out only in the lower bits of samples or color components. In addition, pirates can easily destroy embedded information, for this it is enough to write pseudo-random binary numbers in all the lower bits of the samples. This will not degrade the quality of the multimedia product, but it will destroy the hidden message (attachment, label, logo).

There is a method of hiding digital data in graphic or sound files [6], according to which the hidden file and the container file are divided into many identical data blocks in size, the data blocks of the container file are numbered (form indices), each data block of the hidden file is compared with the blocks data of the container, look for the most similar data blocks in the container, determine the index numbers of the most similar data blocks in the container for each data block from the hidden file and hide all the received indexes in ml the last bit bits of the container.

The disadvantage of this method is the impossibility of hidden text transmission, as well as severe distortion of hidden audio and image files. The occurrence of distortions is due to the inability to find completely identical blocks in the container file and in the hidden file.

A known method of steganographic transmission of information in a TCP / IP network [1], in which additional information is covertly transmitted by changing the length of TCP / IP packets. All transmitted packets in accordance with the secret key are divided into information and masking. Masking packages are not used to attach additional information. The length of these packets is changed randomly. Masking packets are used only to complicate cryptanalysis. Using camouflage and information packets, camouflage (meaningless) text is transmitted over the network. Additional information is hidden in the length of only information packets.

In the considered technical solution, the effective idea of spatio-temporal spraying of information was used [2]. In accordance with this idea, the transmitted information is crushed into small parts and sprayed in space and time. In this case, the introduction of additional information does not occur in all packets, but only in information packets and not in all bits of the binary number that determines the length of the packet, but only in bits that also determine the secret key.

The described technical solution cannot be used directly (without significant modifications) for embedding additional information in sound files, which is necessary, for example, to protect copyright. It is intended for the introduction of additional information into packets and transmissions in networks with the TCP / IP protocol.

The closest technical solution (prototype) is the method described in the American patent [3].

According to method [3], one or more least significant bits in each sample of the sound file are replaced in a predetermined manner with additional information bits, the number of bits embedded in one sample being changed depending on the instantaneous absolute value of the sample.

The purpose of the invention [3] is to increase throughput, that is, to increase the number of bits hidden in one sample. In the case when the digital readout is large, the implementation is carried out using not only the least significant, but also the higher bits of the sample.

As a result of using this invention, additional information is not embedded in the container when the level of the analog audio signal is low, and with increasing level of the analog signal, the number of bits used in the sample increases.

The disadvantage of this invention is the rigidity of the algorithm used, the introduction of information into the sound file is monotonous and predictable. Knowing the implementation algorithm, it is not difficult for cryptanalysts to determine the samples in which additional information is recorded. The definition of the samples used allows you to either extract secret information or destroy it (in the case of watermarks).

The technical result of the proposed method is to increase the cryptographic strength.

The essence of the proposed method of steganographic embedding of additional information in samples of digital audio signals is that encrypted hidden information in binary form is bitwise embedded in samples by replacing the sample bit with a hidden bit of additional information, only samples with a large absolute value of the digital code are used, the implementation is not only in the lower bits of samples, but also in the higher bits, moreover, sound channels used for the introduction of additional information, bit numbers emplov, the interval between samples is used for introduction, and the sequence of the introduction of additional information bits in the samples selected in accordance with a pseudo-random secret key and with the psychophysical characteristics of human hearing.

The specified result is achieved by the fact that the embedded information is converted into binary code, the resulting binary sequence is encrypted with a cryptographic cipher and bitwise embedded in digital samples of sound signals (samples). The choice of samples for embedding, the numbers of bits used in the sample of the digits, the order of embedding the cryptogram in the samples, as well as the selection of the sound channel is done using a pseudo-random key, which is known only on the transmitting and receiving sides.

Confirmation of the high cryptographic strength of the proposed method of covert information transfer are the following considerations.

To extract information from a container without a key, a cryptanalyst must solve such problems.

1. Determine in which channels there are samples with embedded information.

2. Determine which samples were embedded (divide the samples into informational and masking).

3. Determine which categories of used (information) samples were implemented.

4. Extract the bits from the information samples in the desired sequence.

5. Decrypt the resulting sequence of extracted bits.

Such a task is non-trivial and computationally complex. Provided that the keys are cryptographically reliable, this task is unsolvable at a given level of science and technology in an acceptable amount of time.

The solution to the problem is complicated by the fact that the choice of sound channels (left-right, first-fifth-third ...), the numbers of the samples used, the numbers of the digits, the sequence of the introduction of hidden information in the samples is determined according to the pseudo-random law. Two adjacent embedded bits are in different sound channels, in different samples, and the used discharge of the sample is unpredictable. Pre-encryption of embedded information further complicates cryptanalysis. Performing bit mixing is equivalent to additional permutation encryption.

In addition, the use of only samples with a large code value (with a large amplitude of the analog signal) complicates the statistical analysis of the sound file and the identification of additional information used to transmit.

The implementation of the invention

Using the proposed method, it is possible to organize a hidden transmission (or storage) of various types of information: text, graphic, sound. In any case, the hidden information must be converted to digital form before implementation.

The most vividly illustrate the implementation of the proposed technical solution is possible by the example of a hidden embedding of a text (message) into a digitized sound signal.

It is advisable to encrypt the embedded text first and then convert it to binary code (this can be done in the reverse order: convert the text to binary code and then encrypt it). A large number of encryption algorithms are known, some of which have become national standards [4]. Using one of the cryptographic ciphers allows you to create multi-level protection of transmitted information [2].

The sound file (container) in which the message will be placed is selected. For definiteness, we assume that a sound file of the WAV format is selected as a container.

In accordance with the secret key, one of the channels of the sound file is selected for implementation. We assume that stereo sound is currently being used and the left sound channel is defined for the implementation with a session key.

As you know, the digitized signal is a sample (samples, samples) obtained as a result of analog-to-digital conversion. Samples are most often 8, 16 24-bit or 32-bit. The sampling frequency can also be different, for example, 11.025 kHz, 22.05 kHz or 44.1 kHz.

In FIG. 1 schematically shows a diagram of some digitized sound.

In the WAV sound format, with a sound depth (bit depth) of more than 8, samples represent positive and negative integers. The samples in the figure are conventionally shown as multi-bit binary numbers (the least significant bits are located near the time axis). Samples located in the first quadrant are represented by positive numbers. Samples from the fourth quadrant are represented by negative numbers. The height of the samples is proportional to the intensity of the sound.

In the following figure (Fig. 2), the samples into which the implementation of additional information is performed are shaded. Such samples can be called informational. The remaining samples can be called masking (not used), since they do not have hidden information.

The separation of samples into information and masking occurs in accordance with the secret key.

In FIG. 3 the dark shading shows the categories of samples into which the hidden information is embedded.

Thus, for the technical implementation of the proposed method, four pseudorandom number generators are required.

Using the first pseudo-random number generator (PRNG) on the transmitting side, an audio channel is selected for implementation. For example, for a stereo system, the sequence generated by PRNG 1 can be: left-left-right-left-left-right-right ... For a six-channel sound file, the sequence of channels used can be: 3-4-2-2-6- 1-2-5-5-6-3-1 ...

Using the PRNG 2, all files are divided into informational and masking ones. It is most convenient to generate pseudo-random numbers that show how many masking samples are located after the information sample. For example, the number 0 indicates that two information samples follow each other. The number 1 says that there is one masking sample between the information samples. The number 2 indicates that between the information samples there are two “empty” samples, etc.

The PRNG generator 3 selects the bits of samples for the introduction of additional information in the selected sample.

The PRNG generator 4 determines in which sequence the bits of the additional information are embedded in the samples.

The order of operation of the device on the transmitting side will be as follows.

The synchronization device, after the formation of the cryptosystem start operation marker, starts the PRNG 1, which determines which sound channel the first injection is made into. Then, the PRNG 2 is turned on, which selects the number of the information sample. Finally, PRNG 3 is triggered, which indicates to which bits of the sample the implementation of additional information occurs. The PRNG generator 4 shows in what sequence the bits of the secret message are embedded in the samples.

Naturally, four similar PRNGs with the same initial settings and in the same sequence should work on the receiving side. This synchronizes the selection of the audio channel, sample number, sample discharge, and the recording order of the embedded bits.

This is the main idea of the implementation of the method.

However, there are two more important features of the proposed method.

The first feature (detail) is that samples with a small code value (in terms of an analog signal, these are samples corresponding to small amplitude values) are excluded from use. In other words, in samples with a small absolute value of the code, the introduction of additional information never occurs.

FIG. 4 illustrates this point.

Based on the analysis of the psychophysical characteristics of human hearing, a threshold level of Npor is selected. For the introduction of additional information, only samples whose magnitude exceeds Npor in magnitude are used. The PRNG generator 2 only counts those samples whose value exceeds Npor.

The second important feature of the proposed method is that the embedded bits of additional information are subjected to mixing.

Assume that the embedded information consists of separate bytes. The byte structure is shown in the figure (Fig. 5).

The least significant digit is marked with the number 8, and the highest one is marked with the number 1. Writing the shown eight bits to the samples can be done in different sequences. For example, 1-2-3-4-5-6-7-8, 8-7-6-5-4-3-2-1, 1-3-5-7-2-4-6-8, 1-8-2-7-3-6-4-5, etc.

The PRNG generator 4 determines in which sequence the bits of the additional information are embedded in the samples. In essence, using PRSP 4, additional encryption is implemented using the permutation method.

In the claims there is a phrase: “sound channels used for embedding additional information, sample bit numbers, the interval between the samples used for embedding, and the sequence of embedding additional information bits in the samples are selected in accordance with the secret pseudo-random key”. This phrase should be understood as follows: on the transmitting and receiving sides there is a PRNG of the same design, but the initial settings for these generators are secret, it is these initial settings that are the keys. They are known on the transmitting and receiving sides.

Let us explain how additional information is recorded in a digital readout. Masks are used to write information to the desired sample bit. For example, to write 0 to the least significant digit of an eight-bit sample, the bitwise logical multiplication of the sample by the mask 11111110 is performed.

To write a logical unit to a sample, you need to perform the disjunction operation 00000001. Similarly, the sample is introduced into the second and subsequent digits: 111111101 and 00000010 ...

When familiarizing yourself with the operation of the GPSCH2 and GPSCH 3 generators, you might get the impression that all the samples and bits of the information samples are randomly selected. In fact, there are limitations that are determined on the basis of experimental studies of the psychophysical characteristics of human hearing. And these restrictions are inherent in the design of pseudorandom number generators.

In FIG. Figure 6 shows the dependence of the audibility of the sound signal f (x) on the number of the category in which the additional information was introduced. 16-bit samples were used, and the numbering of the digits was carried out from the highest digits to the lowest digits (the senior digit was serial number 1, and the lowest digit was number 16).

In the experimental studies, 20 students participated, who evaluated the sound volume of the presented 555 files using a five-point system (from 0 to 4). The files were a record of "silence" and differed from each other in that the unit was recorded in different samples and different digits. The highest volume of sound corresponded to a score of 4, and a file without an attachment (“complete silence”) corresponded to a score of 0.

FIG. 6 shows that the implementation of a unit of 16, 15, 14 and 13 categories by listeners is not perceived even when reproducing “silence” (pauses between musical works). Obviously, with an increase in the volume of the sound signal, the introduction of information into the older bits becomes permissible.

It was experimentally established that the organoleptic perception of sound depends on the frequency of appearance of loud sounds. The presence of rare even loud sounds by a person is not perceived due to the inertia of hearing.

In FIG. 7 shows how the audibility (sound volume) of the presence of attachments depends on a change in the interval between information samples (implementation was carried out in bit 12).

The figure shows the nonlinear frequency dependence of the auditory system of people. The highest sensitivity is observed at a frequency of about 3.4 kHz. This is in good agreement with previously published data [5]. It can also be seen from the figure that if 60-70 masking samples are located between the information samples, the sound signal is not heard even when the sound file is inserted into the fragments with complete “silence” recording.

The parameters of the PRNG 2 and PRNG 3 generators are interconnected: the larger the interval between the information samples, the more information can be embedded in the older bits. When choosing the parameters of PRNG 2 and PRNG 3, it is recommended to focus on the data in the following table.

Table 1 should be interpreted as follows.

If you select data from the second column for tuning, the PRNG 3 must randomly generate the numbers 13, 14, 15 and 16, that is, the second row of the table indicates the highest digit of the sample allowed for implementation. Recall that the low-order bit is numbered 16. PRNG 2 should generate a pseudo-random number indicating the number of masking samples located between two adjacent information samples. In this case, this number should be 20 or more. The upper boundary of the random numbers generated by the PRNG 2 is selected for reasons of necessary cryptographic strength and performance. The larger the upper limit of numbers, the higher the cryptographic strength, but less performance, that is, less additional information is transmitted per unit time.

The rest of the data in the table is treated in the same way. For example, the numbers in column 3 say that the numbers of the numbers of the used digits 11, 12, ... 16 and the number of masking samples from 100 to, say, 200 should be formed pseudo-randomly.

It should be noted that the data in Table 1 are compiled taking into account the use of a fragment of the sound picture with the minimum sound intensity (worst case). This ensures no audibility of attachments in real sound files. Moreover, the claimed method involves the complete exclusion of the use of samples with a small value. For these purposes, a comparator with a set value of Npoor is used.

It is clear that the hardware implementation of this method can be different: in the form of a specialized device (consisting of microcircuits of medium and high degree of integration) and in the form of a computer (microprocessor) with the corresponding software.

The preferred option is to implement a cryptosystem using a computer (hardware and software implementation of the proposed method).

Consider a block diagram of an algorithm for implementing additional information on the transmitting side (Fig. 8). On the receiving side, the order of operations will be slightly different.

In accordance with the flowchart of the algorithm, after entering the initial data, it is necessary to encrypt the message to be hidden, convert the cryptogram to binary code and split the cryptogram into bytes. In the general case, it does not matter what blocks (groups) of bits the cryptogram is divided into. Blocks can be formed of eight bits (from bytes), of ten, sixteen, etc. bits. The length of the block does not matter. Changing the length of the group will not lead to a change in the structure of the flowchart of the algorithm, only the values of the variables k and d will change. The number k determines the number of blocks in the cryptogram. The number d determines the number of bits in each block.

After breaking the cryptogram into blocks, the procedure for counting the number of processed blocks is organized, for this, the variable m = 1 is first initialized and the counter m = m + 1 is set. Branch block k = m? intended to determine when completion of the implementation of additional information in the container. When the number of blocks embedded in the container becomes equal to the number of blocks available in the message, the program ends.

The first in the transmission is triggered by PRNG 4, which determines the order of permutations of bits in the selected block (for example, in byte). In accordance with the received number, bits are mixed in the block. Note that on the receiving side, the PRNG 4 is triggered by the last of the pseudo-random number generators. After the PRNG 4 on the transmission, the PRNG 1, PRNG 2 and PRNG 3 are triggered, with the help of which sound channels for injection, samples and samples are selected. In the general case, it is permissible to implement not one sample bit, but several. In this case, it is advisable to use an additional PRNG. Embedding is performed only in samples whose code exceeds N then.

The variable initialization block i = 1, the counter i = i + 1, the branch block i = d? serve to determine whether all bits of one block are embedded in an information sample. The symbol d denotes the number of bits (bits) in the block.

After the introduction of all message blocks into the sound container, the sound file containing the attachment is output.

To reduce the size of the picture and to clarify the idea, some blocks are not shown in FIG. 8. For example, operations in the PRNC block 4 need to be understood as follows: a pseudo-random number is generated and bit permutations are performed. The PRNG block 1,2,3 needs to be interpreted as follows: three random numbers are generated and, in accordance with them, the sound channel, sample, and sample discharge are selected. The sample cutoff block with a low level is not shown. The task is trivial, but the introduction of these blocks in the block diagram of the algorithm greatly obscures the main ideas.

It is of interest to estimate the size of the container needed to implement additional information.

The digitized three-minute monophonic piece of music with a sampling frequency of 44.1 kHz contains more than 7 million samples. About a third of the samples should be excluded due to their low value. Since in accordance with the claimed method, the introduction is not carried out in all samples of large size (but, for example, in every hundredth suitable sample), the number of information samples will be a little more than 20 thousand. If we consider that the implementation is carried out only in one category of the information sample, then the amount of additional information embedded in the container will be approximately 20 Kbytes. This value can be increased 2 ... 25 times by reducing the number of masking samples (reducing the number of masking samples from 100 to 20) and the introduction of several bits in each information sample (2 ... 5). In addition, the amount of information that can be hidden can be increased several times when using multi-channel audio formats (for example, stereo).

However, it is obvious that the inventive method should be used mainly for the secret transfer of text, keys, passwords, credit card numbers or small graphic files (for example, the logo of the copyright holder). It is clear that any information presented in digital form can be transmitted.

Positive features of the technical solution

The main advantage of the proposed method is the created ability to covertly transmit additional information that can be extracted only with a secret key. High cryptographic strength is achieved by the fact that the enemy does not know in advance in which sound channel, in which sample and in which category of the sample the implementation was made. Additional levels of cryptographic protection are created due to the fact that preliminary encryption of the message and pseudo-random permutation of bits in the data block are performed. At the current level of development of computer technology, it is impossible to extract confidential information without knowing the key.

When using the claimed invention for copyright protection, the company logo can be recorded in a musical work several times. At the same time, dividing the product into parts does not allow to hide authorship, since the logo will be present in all fragments (in all subsets of the multimedia product).

The developed technical solution allows you to choose the desired ratio between cryptographic strength and performance by adjusting the pseudo-random number generators responsible for the choice of sound channels, information samples and discharges in samples.

To obtain a stegosystem with maximum cryptographic strength (steg resistance), the introduction should be made only in the lower bits of the samples, maximizing the number of masking samples (spraying the attachment throughout the container) and, if possible, less often use each sound channel.

The maximum performance of the stegosystem is achieved by introducing several bits of each information sample, the greatest reduction in the number of masking samples and the intensive use of each sound channel.

The parameters of the PRNG provided in the description of the invention are obtained on the basis of experimental studies of the psychophysical properties of people and guarantee the inability to detect attachments in a sound container by ear.

The claimed technical solution allows you to reliably protect copyrights, since it is impossible to remove additional information in high-order samples and maintain an acceptable sound quality. If the recording was carried out only in the lower digits of the samples, then it would be possible to fill them with pseudo-random ones and zeros.

Literature

1. Orlov V.V., Alekseev A.P. A method of steganographic transmission of information in a TCP / IP network. Patent 2463670. Application 2010125304/08 (035921). Application submission date 06/18/2010. IPC G09C 1/00, H04L 9/00. Positive Res. March 23, 2012

2. Alekseev A.P., Makarov M.I. The principles of multilevel information protection // Infocommunication technologies, volume 10, No. 2, 2012. Pp. 88-93.

3. US patent US 4750173 A. Method of transmitting audio information and additional information in digital form.

4. Panasenko S.P. Encryption algorithms. Special reference. - St. Petersburg: BHV-Petersburg, 2009 .-- 576 p.

5. Aldoshina I.A. Electroacoustics and sound broadcasting: textbook. manual for universities / I.A. Aldoshina, E.I. Vologdin, A.P. Efimov, G.P. Katunin, L.N. Katznelson, Yu.A. Kovalgin, A.A. Fadeev; ed. Yu.A. Kovalgin. M .: Hot line - Telecom, Radio and communications, 2007.872 s.

6. US patent US 6023511 A. Cryptosystem for encrypting digital image or voice file.

7. Agranovsky A.B., Devyanin P.N., Hadi P.A., Cheremushkin A.V. Basics of computer steganography. Textbook for universities. - M .: Radio and communications, 2003.152 s.

8. US patent US 5850481 A. Steganographic system.

Claims (1)

The method of steganographic embedding of additional information in samples of digital audio signals, which consists in the fact that the encrypted hidden information in binary form is bitwise embedded in the samples by replacing the sample bit with the hidden bit of additional information, only samples with a large absolute value of the digital code are used, the implementation is performed not only in the lower bits of the samples, but also in the higher bits used to introduce additional information, sound channels, numbers of bits of the samples, the interval between the samples used for embedding, as well as the sequence of embedding bits of additional information in the samples, are selected in accordance with the secret pseudo-random key and taking into account the psychophysical characteristics of human hearing.

Images