RU2288544C2 - Method for embedding additional information into digital images - Google Patents

Abstract

FIELD: cryptography technique; hidden storage and transfer of confidential information through open communication channels; marking of images containing large amounts of additional information.

SUBSTANCE: method for embedding additional information into digital images consists in substitution of separate bits in bytes of initial image. The remaining part of bits is used for correction of final digital image. The initial digital image is separated into bit layers. One of the obtained bit layers which is represented by bit sequence is chosen for writing additional information. Writing of additional information into obtained bit sequence is made using code. During writing of additional information using code bits in the obtained bit sequence, that are located on the margins of all changes of same bit sequences of zeroes and ones, are replaced in accordance to bits of written additional information.

EFFECT: increased volume of embedded information and ensured high tolerance of messages against some steganoanalysis methods.

2 dwg, 4 tbl

Description

The invention relates to cryptographic techniques and is intended for the organization of hidden storage and transmission of confidential information through open communication channels, can also be used for marking images containing large amounts of additional information.

Known methods for steganographic protection of information based on the introduction of additional information in raster images by replacing the least significant bits in bytes of the color representation of the original image with bits of added information [Bender W., Gruhl D., Morimoto N., Lu A., Techniques for data hiding, IBM systems journal, VOL 35, NOS 3 & 4, 1996. RU 2002135272, 07/10/2004].

The main disadvantage of such methods is the low efficiency of information protection, since the changes introduced by them during the implementation of information lead to a distortion of a number of statistical parameters of processed images, which can be detected by statistical analysis methods [Andreas W., Andreas P., Attacks on Steganographic Systems. Breaking the Steganographic Utilities EzStego, Jsteg, Steganos, and S-Tools - and Some Lessons Learned, Proceedings of the Workshop on Information Hiding, 1999]. At present, such analysis methods as visual stegoanalysis, analysis of series lengths, analysis of the number of transitions, estimation of entropy, analysis of histograms based on the χ ² criterion are effectively used.

There is also a method of embedding additional information in images, in which to counteract some analysis methods when recording information, only part of the least significant bits in bytes of the color representation of the original image are used, the remaining bits are used for subsequent correction of the most important statistical parameters [Niels Proves, Defending Against Statistical Steganalysis , Proceeding of the 10 USENIX Security Symposium, 2001, P.323-335].

The disadvantage of this method is that corrective changes lead to additional distortion of the container and reduce the bandwidth of the stegochannel due to the use of part of the free space for corrective information. In addition, corrective corrections aimed at counteracting any of the methods of statistical analysis by correcting one of the parameters can lead to even more distortion of other parameters and, as a result, to more reliable detection of the covert channel by other analysis methods. In particular, the distortions introduced by the introduction of additional information and corrective changes when using this method turn out to be easily noticeable during visual analysis of bit slices of the obtained image.

The disadvantages of the above methods of embedding information in bitmap images should also include the fact that only the least significant bits of the bytes of the color representation can be used to embed information, which does not allow the full use of the entire amount of image data.

The task this invention is directed to is to increase the level of protection of confidential information when it is hidden and transmitted over open communication channels, if raster photorealistic images are used as containers, and also to increase the relative amount of hidden information through the use of medium and high digits information bytes of the image.

This is achieved due to the fact that for the introduction of information, those bits of the original image are selected whose behavior is the least predictable and whose change does not lead to noticeable distortions of the original image. As such bits, the encoder selects the bits located at the boundaries between different regions of the image, while the search for the corresponding bits in the image is performed automatically immediately during the recording of additional information. A bit sequence is extracted from the processed image, which is fed to the input of the encoder or decoder, while recording or retrieving information is carried out only in cases where a portion of the bit sequence containing a transition from 0 to 1 or from 1 to 0 falls into the viewing window of the encoder or decoder In the process of embedding information, both the bits belonging to the selected bit sequence and the remaining bits of the image can be changed. In this case, part of the bits can be used to compensate for the introduced distortions both in the process of recording information and in the subsequent correction of the received image.

The technical result is to increase the amount of embedded information through the use of both the least significant and the most significant and significant bits of bytes of the color representation of the image points, while maintaining the visual quality of the image. The latter is possible due to the fact that in photorealistic images the behavior of points located on the border between different regions of the image obtained by digital conversion depends on a large number of random factors and is difficult to predict. Using a special code, in turn, allows you to save most of the statistical parameters of the image with sufficiently high accuracy and to minimize the number of additionally controlled parameters and corrective changes, thanks to which it is possible to successfully resist a number of statistical analysis methods. In addition, the fact of the presence of extraneous information cannot be established during the visual analysis of bit slices of the received image.

The essence of the method is illustrated by the example of processing by the encoder and decoder the sequence of bits of the original image of figure 1, obtained in accordance with the scheme in figure 2.

The formation of the bit sequence is carried out according to the scheme of sequential image processing shown in figure 2.

In the case of processing a color image, it must be presented in the form of three monotonous images corresponding to the decomposition into red, green and blue components, black and white images are not subjected to such decomposition. Further, since one byte is enough to store information about the color of each point of the received images, they are decomposed into eight bit layers. To embed information, one of the obtained layers is selected, which is presented in the form of a bit sequence. Subsequent recording of information is carried out by changing the individual bits of the obtained sequence, while all the underlying layers can be used for additional correction or for recording information in the next step.

It should be noted that in the bit sequences of the source images there are quite long series of the same bits. In the case of recording information by directly replacing all bits of the original sequence with the bits of the message to be hidden, the series of identical bits will be destroyed, and the bit sequence itself will lose touch with the original image (correlation dependencies are violated). In this case, the fact of recording additional information can be easily established both as a result of a simple statistical analysis of the image by bit layers, and visually. To ensure the secrecy of recording information and minimizing visual distortions, in the process of recording additional information in a bit sequence, it is necessary to ensure the safety of long series of the same bits. In addition, in order to counteract some statistical analysis methods, it is also necessary to preserve the initial ratios of the number of zeros and ones in the bit sequence and the number of transitions from one to zero and from zero to one to the total length of the sequence.

The most fully described requirements are met by the method of recording additional information in a bit sequence, which, during the recording of information, changes only the bits located at the transition boundaries of sequences of identical bits. So, for example, in the sequence 0000 01 1111 ... either only the fifth or only the sixth bit can be changed, respectively, as a result, two versions of the changed sequence can be obtained:

0000111111 ... (change of the fifth bit)

0000001111 ... (sixth bit change)

As you can see, such changes in the bit sequence do not lead to the destruction of long series of identical bits and the appearance of new transitions, as in the case of changing other bits of the sequence or simultaneously changing the fifth and sixth bits:

0101011011 ... (change of the second, fourth and eighth bits)

0000101111 ... (change of the fifth and sixth bits)

The criteria for extracting information from a bit sequence can be the lengths of sequences consisting of the same bits and the position of the boundaries (position of elements at the beginning and at the end of sequences of identical bits). Using as a criterion the positions of the boundaries of sequences of identical bits is more preferable due to the greater information capacity. Below is an example that demonstrates how, by replacing bits located at the junction boundaries of bit sequences of zeros and ones, information can be recorded and retrieved.

An example of recording information in a bit sequence. Suppose you have the original bit sequence C, and you want to record the message E = 10010110. A simple decoder considers a bit sequence in pairs and extracts information according to the following rules.

1. If the pair contains 1 and 0, then, depending on their order, the decoder writes 1 (corresponds to 10) or 0 (corresponds to 01).

2. If a pair consists of only ones or only zeros, then it is skipped.

The decoder code table corresponding to these rules is represented by table 1.

Table 1. entrance Result 00 - 01 0 10 one eleven -

Knowing the rules for extracting information by the decoder, based on the original bit sequence C, by changing the boundary bits in accordance with the bits of the recorded message E, a new bit sequence S = С⊕Е can be formed. The bits of the original bit sequence that have changed are marked in bold.

C = 11100000000111001101111011011001001110011 ...

E = 10010110 ...

S = C⊕E

S = 11 10 00 00 00 01 11 01 11 00 11 10 11 01 10 10 00 11 00 01 1 ...

Y = x1xxx0x0xxx1x011xxx0 ...

E '= 10010110 ...

Based on the bit sequence S according to the code table, the decoder restores the message E 'corresponding to the message E recorded in the sequence S: S → Y → E'.

A feature of the proposed method for recording information in a bit sequence is to preserve the original entropy of the bit sequence, that is, information is recorded not by adding new additional information to the sequence, but by changing the information already present in it. This recording feature is related to the fact that in the above example, during the recording of information, only five bits underwent a change in the original sequence. This is explained by the fact that part of the information that was originally present in the bit sequence coincided with the recorded information.

It is proposed to write information in a bit sequence using a special code. Its application allows you to automate the process of recording information and reduce it to a simple replacement of four-bit vectors formed from successive bits of the original sequence by the vectors of the replacement table. The process of recording information for this case is presented in figure 1. At each moment of time, the encoder scans the sequence of bits of the container in fours, while only the middle two bits undergo a change. The bits on the left and on the right allow you to more accurately determine the boundary of transitions and exclude situations where the boundary is between the pairs in question. The shift of the encoder viewing window at each step is two positions, this allows information to be recorded in all transitions of sequences of identical bits. The proposed encoder replaces the elements of the bit sequence in accordance with the recorded data according to the code table. The code table consists of two columns containing replacement vectors for zero and a unit of recorded information, respectively. Rows of the table correspond to all possible four-bit vectors of the bit sequence. At each step, the encoder viewing window will highlight a four-bit segment of the sequence (vector) and, in accordance with the code table for the current information symbol (data bit), it will replace it with a vector from the code table.

The encoder code table is presented in Table 2. The proposed table is optimal, and its use leads to minimal distortion, however, in some implementations, the use of other code tables is permissible. The code described in Table 2 is basic, does not introduce additional corrective changes, and allows using up to six lower-order bit layers for recording information.

Table 2. Input (4-bit vector) Zero recording Unit record 0000 - - 0001 0011 * - 0010 0010 0100 0011 0011 0111 * 0100 0010 0100 0101 0011 0101 0110 0010 * 0100 * 0111 0011 * 0111 1000 1000 1100 * 1001 1011 * 1101 * 1010 1010 1100 1011 1011 1101 1100 1000 * 1100 1101 1011 1101 1110 - 1100 * 1111 - -

An image with additional information embedded in it is obtained by performing operations that are inverse to operations carried out during the formation of the bit sequence, but the last is not the original, but the changed bit sequence.

Information extraction is carried out as follows. Similar to the procedure for recording information from an image according to the scheme in figure 2, a bit sequence is formed. The decoder analyzes this sequence in triples and extracts information in accordance with its code table (table 3). The offset of the viewing window of the decoder at each step is two positions (figure 1).

Table 3. entrance Result 000 - 001 0 010 one 011 one one hundred 0 101 0 110 one 111 -

Making corrective changes.

The substitution options marked with asterisks in Table 2 lead to a change in the ratio of the number of ones and zeros, but their influence on the overall statistics is mutually compensated and the maximum deviation does not exceed 1.45%, and the maximum distortion of the number of transitions is only about 0.25%. Such a deviation is not detected by modern means of stegoanalysis, however, in order to counter more advanced analysis tools, these parameters can be maintained with almost absolute accuracy, due to the introduction of additional corrective changes. To do this, additional states are added to the rows of the base code table 2. A possible code table is presented in table 4.

Table 4. entrance Zero recording Unit record Correction 0000 - - - 0001 0011 - - 0010 0010 0100 0110 0010 0011 0111 0101 (entry 1) 0100 0010 0100 - 0101 0011 0101 0001 0110 0010 0100 - 0111 0011 0111 - 1000 1000 1100 - 1001 1011 1101 - 1010 1010 1100 1000 1011 1011 1101 - 1100 1000 1100 1010 (entry 0) 1101 1011 1101 1001 1110 - 1100 - 1111 - - -

Using a modified code table in the coding module allows you to make additional corrective changes as necessary directly during the recording of information, while the introduction of corrective changes in some cases may be accompanied by a record of information. As, for example, in lines 4, 13 of the given code table. Moreover, this approach allows you to save all the basic statistical characteristics of the original sequence with great accuracy, without subjecting the decoder to change.

During the recording of information in one of the bit layers, all the underlying layers can be used to make corrective changes. In this case, only bits corresponding to those image points that were changed during the recording of information can be subjected to change. This condition is necessary to maintain the visual quality of the image and the possibility of subsequent use of the lower layers for recording information. In addition, it allows you to record information and the necessary correction in one pass. So, for example, the use of this approach allows you to accurately preserve the histogram of the color distribution of the original image.

As noted above, the basic code allows the use of up to six bit layers for recording information. At the same time, if a black-and-white image is used to record information, then due to the introduction of additional compensating changes in the underlying layers, in some cases it is possible to use higher layers to record information. In the case of a color image, matching of all color components is additionally required.

The tests performed showed the high stability of the basic method of introducing additional information in the image against visual analysis, a number of statistical methods for image analysis, as well as histogram analysis. During the tests, the relative amount of embedded data in some cases was 1 / 3-1 / 2 of the data volume of the image itself, while maintaining the visual quality of the latter. At the same time, the use of the proposed modifications makes it possible to further increase the steg resistance of the method and use additional bit layers for recording information. To ensure maximum persistence, the information bit sequence must have a uniform distribution. The most effective is the use of the proposed method of hiding information in conjunction with the compression and block encryption algorithms applied sequentially to a hidden message.

List of figures:

1. FIG. 1: An example of processing a sequence of image bits by an encoder and a decoder.

2. Figure 2: The procedure for obtaining the bit sequence from a color image presented in digital form.

Information sources

1. Bender W., Gruhl D., Morimoto N., Lu A., Techniques for data hiding, IBM systems journal, VOL 35, NOS 3 & 4, 1996.

2. Patent RU 2002135272, 07/10/2004.

3. Andreas W., Andreas P., Attacks on Steganographic Systems. Breaking the Steganographic Utilities EzStego, Jsteg, Steganos, and S-Tools - and Some Lessons Learned, Proceedings of the Workshop on Information Hiding, 1999.

4. Niels Proves, Defending Against Statistical Steganalysis, Proceeding of the 10 USENIX Security Symposium, 2001, P.323-335.

Claims (1)

The method of embedding additional information in a digital image, which consists in replacing individual bits in bytes of the original digital image and using part of the remaining bits to correct the final digital image, characterized in that the original digital image is laid out in bit layers, one of the received bit data is selected for recording additional information layers, which are presented in the form of a bit sequence, the recording of additional information in the received bit sequence carried out they are determined using the code, and when recording additional information using the code in the obtained bit sequence, the bits located at the boundaries of all transitions of the same bit sequences of zeros and ones are replaced in accordance with the bits of the additional information recorded, the remaining bits of the underlying bit layers are used for correction if necessary source image or to record other additional information.

Images