JP4920045B2 - Method for identifying marked images based at least in part on frequency domain coefficient differences - Google Patents

Description

The present application relates to classifying or identifying content such as, for example, marked images.

Digital data hiding technology is a field that has been actively studied in recent years, and various data hiding methods have been proposed. Some methods aim at content protection and / or authentication, while others aim at secret communication. Data hiding classified as the latter is referred to herein as steganography.

The subject matter of the present invention is set forth with particularity in the appended claims and is claimed explicitly. However, the claimed subject matter will be best understood by reference to the following detailed description and the accompanying drawings, as well as its purpose, features and / or advantages, both in terms of the mechanism and method of operation.

DETAILED DESCRIPTION In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of claimed subject matter. However, one of ordinary skill in the art appreciates that the subject matter recited in the claims can be practiced without the specific details. In other instances, well-known methods, procedures, components, and / or circuits have not been described in detail so as not to obscure the claimed subject matter.

In the detailed description that follows, an algorithm and / or symbolic representation of the manipulation of data bits and / or binary digital signals stored in a computing system (such as a computer and / or computing system memory) is described. There is a part. These algorithmic descriptions and / or representations are techniques used by those skilled in the data processing arts to convey their work to others skilled in the art. An algorithm is herein and generally considered a self-consistent sequence of operations and / or similar processes that yields the desired result. The above operations and / or processes may involve physical manipulation of physical quantities. These physical quantities are typically, but not necessarily, in the form of electrical and / or magnetic signals that can be stored, transferred, combined, compared, and / or otherwise manipulated. It may be convenient to refer to these signals as bits, data, values, elements, symbols, characters, terms, numbers, and / or numbers, primarily in terms of general usage. It should be understood, however, that these and similar terms are all associated with the appropriate physical quantities and are merely convenient labels. Unless stated otherwise, the description using terms such as “processing”, “arithmetic”, “calculation”, and / or “determining” is used herein to be a computing platform, as will be apparent from the following description. Process data represented as physical electronic quantities and / or magnetic quantities and / or other physical quantities in a processor, memory, register, and / or other information storage device, information transmission device, and / or information display device It is understood that it refers to operations and / or processes performed by a computing platform (such as a computer or similar electronic computing device) that transforms.

Due to the widespread use of JPEG images, steganographic tools for JPEG images have increased in recent years, among which model-based steganography (MB), F5 and OutGuess are the latest. However, the development of new tools for identifying images containing hidden data remains desirable. According to the claimed subject matter, one embodiment described herein is at least a statistical moment derived at least partially from an image 2-D array and a JPEG two-dimensional array. Including partially based methods. In this particular embodiment, a first order histogram and / or a second order histogram may be used, but the claimed subject matter is not limited in scope in this respect. For example, in other embodiments, for example, higher order histograms may be utilized. However, in this particular embodiment, the moments of the two-dimensional characteristic function are also used from these histograms, but again, other embodiments are not limited in this respect. For example, a higher order moment may be used.

The widespread use of the Internet is accelerating due to the widespread use of computers. As a result, millions of videos are flowing on the Internet every day. In recent years, exchange of JPEG (Joint Photographic Experts Group) images has been performed more and more frequently. Many steganographic techniques for manipulating JPEG images have been released and are publicly used. Most of this type of technology appears to embed hidden data by changing 8x8 block discrete cosine transform (BDCT) coefficients in the JPEG domain. Among steganographic techniques, the recently published schemes OutGuess, F5 and model-based steganography (MB) appear to be the latest. See the following references: Provos, “Defending Against Statistical Steganization,” 10th USENIX Security Symposium, Washington DC, USA, 2001; Westfeld, “F5 a steganographic algorithm: High capacity de- fect better stegansis,” 4th International Workshop on Information-Hidging, Pittsburgh; Salee, “Model-based steganography,” International Work-shop on Digital Watermarking, Seoul, Korea, 2003. OutGuess embeds the data to be concealed using the redundancy of the cover image. Here, the cover image refers to content in which hidden data is not embedded. For JPEG images, OutGuess attempts to store statistics based at least in part on the BDCT histogram. In addition, OutGuess identifies redundant BDCT coefficients and embeds data in these coefficients to reduce the effects of data embedding. In addition, it tries to preserve the original BDCT histogram by adjusting the coefficients that are not embedded with data. F5, an extension of Jseg, F3 and F4, uses the following techniques: straddling and matrix coding. Stradling distributes the message as evenly as possible on the cover image. Matrix coding tends to improve embedding efficiency (defined herein as the number of embedding bits per change in BDCT coefficients). MB embedding attempts to correlate embedded data with a cover medium. This divides the cover medium into two parts, models the distribution parameters of the second part given the first part, and encodes the second part using the model and the embedded message Implemented by combining the two parts to form a stego medium. Specifically, a JPEG BDCT mode histogram is modeled using a Cauchy distribution, and embedding attempts to keep the low accuracy histogram of the BDCT mode unchanged.

Many steganalysis methods have been proposed for detecting hidden information in stego images. A generalized steganalysis method using higher order statistics has been proposed by Farid. See the following references: Farid, “Detecting hidden messages using high-order statistic-tick models,” International Conference on Image Processing, Rochester, NY, USA, 2002 (hereinafter “Farid”). Decompose the test image into wavelet subbands using orthogonal mirror filters. Higher order statistics are calculated from the wavelet coefficients of the high frequency subbands to form a group of features. Another group of features is similarly shown from the prediction error of the wavelet coefficients of the high frequency subbands. Literature: Y.M. Q. Shi, G .; Xuan, D.C. Zou, J. et al. Gao, C.I. Yang, Z .; Zhang, P.A. Chai, W .; Chen, C.I. Chen, "Steganalysis based on moments of characteristic functions using wavelet decomposition, prediction-error image, and neural network," International Conference on Multimedia and Expo, Amsterdam, Netherlands, 2005 (hereinafter referred to as "Shi et al.") Are described in The method uses the statistical moment of the characteristic function of the test image, its prediction error image, and their discrete wavelet transform (DWT) subbands as features.

However, Fridrich has proposed a steganalysis method designed specifically to deal with the JPEG steganographic scheme. See the following references: Fidrich, “Feature-based steganization for JPEG images and it's implications for future design, steganographic schemes,” 6th Information Hidden Corp. The method uses a set of carefully selected features of relatively small size when detecting images with hidden data generated by Outlooks, F5 and MB over other steganalysis methods as described above. Excellent performance. See the following literature: Kharrazi, H .; T.A. Sencar, N .; D. Memon, “Benchmarking steganographic and steganary techniques,” Security, Steganography, and Watermarking of Multimedia Content, 2005, San Jose, CA.

In recent years, schemes have been developed to detect data concealed by spread spectrum techniques. The scheme uses dependency between pixels and employs a Markov chain model. See the following references: Sullivan, U .; Madhow, S .; Chanderaskaran, and B.C. S. Manjunath, “Steganalysis of Spread Spectrum Data Hiding Exploring Cover Memory,“ The International Society for Optical Engineering, Ill. In this approach, an empirical transition matrix for a given test image is formed. This matrix has 256 × 256 dimensions for one image of a grayscale image with a bit depth of 8. That is, this matrix has 65,536 elements. Since there are many elements, it is difficult to use all elements as features. The author decided to select some of the maximum probabilities along the main diagonal of the matrix along with their neighbors as features, and then randomly select other probabilities along the main diagonal. Naturally, some information loss cannot be avoided by this feature selection process. Furthermore, the above method uses a Markov chain along the horizontal direction, and therefore this approach does not necessarily reflect the two-dimensional nature of the digital image.

There remains a need to distinguish between JPEG images with hidden data and JPEG images without hidden data. One embodiment of the claimed subject matter uses a JPEG two-dimensional array. In this particular embodiment, a JPEG two-dimensional array is formed based at least in part on the JPEG quantized block DCT coefficients. Similarly, in the particular embodiment, differential JPEG 2-D arrays may be formed along horizontal, vertical and diagonal directions, and these differential JPEG 2-dimensional arrays may be applied by applying a Markov process. The second order statistics may be used for steganalysis by modeling. In addition to the use of the differential JPEG two-dimensional array, a threshold processing technique may be applied to reduce the number of dimensions of the transition probability matrix and make the calculation amount of the scheme easier to process.

In this particular embodiment, steganalysis is considered a two-class pattern recognition task. That is, a given image is classified as either a stego image (one that holds hidden data) or a non-stego image (one that does not hold hidden data).
As described above, in the latest steganography methods such as OutGuess and MB, much effort has been made to make detection more difficult by keeping the change in BDCT coefficients due to data hiding relatively small. These methods in particular attempt to keep the change in the histogram of JPEG coefficients relatively small. Therefore, in such a situation, higher-order statistics are desirable as a feature in steganalysis as used in this embodiment. Note that, specifically, the second-order statistics are used in the embodiment, but the scope of the subject matter described in the claims is not limited in this respect.

In this embodiment, a JPEG two-dimensional array is formed. Similarly, a differential JPEG two-dimensional array along various directions is formed. In order to model a differential JPEG two-dimensional array using a Markov random process, a transition probability matrix may be constructed to characterize the Markov process, and features are derived from this transition probability matrix. In the present description, a so-called one-step transition probability matrix is used for the reduced computational complexity, but the claimed subject matter is not limited in scope in this respect. For example, in other embodiments, a more complex transition probability matrix may be used. Although details will be described later, a threshold processing technique can be applied in order to further reduce the amount of calculation.

In this embodiment, features are generated from the DCT block representation of the image, but the claimed subject matter is not limited in scope in this respect. For example, in another embodiment, another frequency domain representation of the image may be used. However, again, in this particular embodiment, it is desirable to consider the characteristics of the JPEG BDCT coefficient.

For a given image, consider a two-dimensional array containing 8x8 block DCT coefficients that have been quantized using the JPEG quantization table but have not been zigzag scanned, run-length coded, or Huffman coded I want to be. That is, this two-dimensional array is the same size as a given image and has a predetermined 8 × 8 block, which is filled with the corresponding JPEG quantized 8 × 8 block DCT coefficients. Has been. Next, apply the absolute values to the DCT coefficients to obtain the two-dimensional array shown in FIG. In the embodiment, the obtained two-dimensional array is referred to as a JPEG two-dimensional array. Although described in detail below, the features in this particular embodiment are formed from a JPEG two-dimensional array.

Without applying absolute value manipulation, the JPEG BDCT quantization factor is either positive, negative, or zero. Normally, BDCT coefficients do not follow a Gaussian distribution, but these coefficients are not necessarily statistically independent of each other. The magnitudes of non-zero BDCT coefficients may be correlated, for example, along a zigzag scan order. Accordingly, there may be a correlation between the absolute values of the BDCT coefficients along the horizontal, vertical, and diagonal directions. This is proved by confirming FIG. 3 shown below. That is, the difference between the absolute values of two adjacent BDCT coefficients that are immediately adjacent to each other (in the horizontal direction in FIG. 3) is remarkably concentrated around zero and exhibits a Laplace-like distribution. The same is observed in the vertical and diagonal directions. That is, as will be discussed later, this particular embodiment utilizes this aspect of the coefficients. However, it will be appreciated that the claimed subject matter is not limited in scope in this respect.

The disturbance caused by data embedding appears more clearly in the prediction error image than in the original image. Therefore, it is desirable to pay attention to the difference between an element and one of its neighboring elements in a JPEG two-dimensional array. Thus, although this particular embodiment uses the following four differential JPEG two-dimensional arrays, the claimed subject matter is not limited in scope in this respect.

A JPEG 2D array generated from a given image

Here, S _u is the size in the horizontal direction of the JPEG two-dimensional array, and S _v is the size in the vertical direction. Then, as shown in FIG. 2, a difference array can be generated as follows:
F _h (u, v) = F (u, v) −F (u + 1, v), (1)
_Fv (u, v) = F (u, v) -F (u, v + 1), (2)
F _d (u, v) = F (u, v) −F (u + 1, v + 1), (3)
F _md (u, v) = F (u + 1, v) −F (u, v + 1), (4)
(Where

And
F _h (u, v), F _v (u, v), F _d (u, v), and F _md (u, v) are differential arrays in the horizontal, vertical, main diagonal, and sub-diagonal directions, respectively. is there).

As suggested above, the distribution of the elements of the difference array may be Laplace-like. Most difference values are close to zero. For evaluation, an image set including 7560 JPEG images having a quality factor of 70 to 90 was accumulated. An arithmetic average of a histogram of a horizontal difference JPEG two-dimensional array generated from this JPEG image set and a histogram of a horizontal difference JPEG two-dimensional array generated from one image randomly selected from this image set, respectively. It shows to Fig.3 (a) and (b). As can be seen from the figure, most elements of the differential JPEG two-dimensional array in the horizontal direction fall within the interval [−T, T] if T is sufficiently large. Average percentage of elements of horizontal difference JPEG two-dimensional array generated from image set that falls within the range [−T, T] when T = {1, 2, 3, 4, 5, 6, 7} Values and variance values are shown in Table 1. 3 and Table 1 both tend to support the notion that the distribution of the elements of the horizontal difference JPEG two-dimensional array is Laplace-like. The same is confirmed for the differential JPEG two-dimensional array along other directions such as the vertical and diagonal directions.

As described above, in the latest steganographic methods such as OutGuess and MB, efforts have been made to keep the change in the histogram of JPEG BDCT coefficients due to data embedding relatively small. Therefore, higher order statistics may be useful in steganalysis of JPEG steganography. In this embodiment, secondary statistics are used so as not to significantly increase the amount of calculation. However, depending on the embodiment and application, it may be desirable to use higher-order statistics exceeding the second order.

In this embodiment, the differential JPEG two-dimensional array is characterized using a Markov random process. Specifically, the Markov process may be characterized using a transition probability matrix. There are so-called one-step transition probability matrices and n-step transition probability matrices. Roughly speaking, the former relates to transition probabilities between two adjacent elements in the difference JPEG two-dimensional array, and the latter is (n−1). ) Related to the transition probability between two separated elements. Although the one-step transition probability matrix as shown in FIG. 4 is used in this embodiment from the balance between the steganalysis performance and the processing amount that can be processed, the scope of the subject matter described in the claims is not limited in this respect.

To further reduce the computational complexity, thresholding techniques may be used, but the claimed subject matter is not limited in scope in this respect. In this embodiment, a threshold value, that is, T in the present specification is used. Therefore, the values are {−T, −T + 1,. . . , -1, 0, 1,. . . , T−1, T}, elements of the differential JPEG two-dimensional array are considered. If the value of an element is greater than T or less than -T, the element is correspondingly represented by T or -T. By this procedure, a transition probability matrix having the number of dimensions (2T + 1) × (2T + 1) is obtained. Of course, again, the claimed subject matter is not limited in scope to using thresholding or to the details of the particular thresholding described above. For example, in other embodiments, the threshold levels are different. However, in this embodiment, the four matrix elements associated with the differential JPEG two-dimensional array in the horizontal, vertical, main diagonal, and sub-diagonal directions are given as follows:

(Where

as well as

Is).

In short, in this embodiment, (2T + 1) × (2T + 1) elements are obtained for one transition probability matrix. Therefore, 4 × (2T + 1) × (2T + 1) elements are generated. Similarly, these may be used as features in steganalysis. In other words, in this particular embodiment, 4 × (2T + 1) × (2T + 1) feature vectors are generated in steganalysis.

As can be seen from the data shown in FIG. 3 and Table 1, T is set to 4 in this example, but the claimed subject matter is not limited in scope in this respect. Therefore, in this embodiment, when the absolute value of an element is larger than 4, the element is replaced with the absolute value 4 with the sign unchanged. The obtained transition probability matrix is a 9 × 9 matrix for the difference JPEG two-dimensional array. That is, there are 9 × 9 = 81 elements per transition probability matrix, ie, 81 × 4 = 324 elements in this particular embodiment. The characteristic configuration in this specific embodiment will be described with reference to the block diagram shown in FIG.

Various techniques can be used to analyze data (referred to herein as features) in various situations. Here, the term “ANOVA” is used to refer to the following processes or techniques: As a result of applying the above processes or techniques, differences resulting from statistical fluctuations are differences resulting from non-statistical fluctuations Processes or techniques that are sufficiently distinguished from each other to correlate, segment, classify, analyze, or otherwise characterize data based at least in part on the application of such processes or techniques. Examples include artificial intelligence technology and processing (such as pattern recognition), neutral networks, genetic processing, heuristics, support vector machines (SVM), etc., which are within the scope of the claimed subject matter. It is not limited.

The claimed subject matter is not limited to SVM or SVM processing, but may be a convenient technique for two-class classification. For example, see the following literature: C.I. Cortes and V.M. Vapnik, “Support-vector networks,” in Machine Learning, 20, 273-297, Kluwer Academic Publishers, 1995. For example, SVM can be used to handle linear and non-linear cases or situations. When linear separation is possible, for example, an SVM classifier is applied to obtain a hyperplane that separates positive patterns from negative patterns.

Thus, for example, Shi et al. Used a neural network, but in this embodiment, a support vector machine (SVM) is used as a classifier. SVM is based at least in part on the idea of hyperplane classifiers, which use a Lagrange multiplier to determine a separation hyperplane that distinguishes positive patterns from negative patterns. When the feature vector is one-dimensional (1-D), the separation hyperplane is given as a point on several axes. SVM can deal with both linear separable cases and non-linear separable cases. Here, the training data pairs are represented as {y _i , ω _i }, i = 1,. . . , L (where

Is a feature vector, N is the number of dimensions of the feature vector, and ω _i = ± 1 for the positive / negative pattern class. Here, an image including hidden data (stego image) is regarded as a positive pattern, and an image not including hidden data is regarded as a negative pattern. However, the scope of the claimed subject matter is not limited in this respect. . In the linear support vector method, one hyperplane H: w ^T y + b = 0, two hyperplanes H ₁ : w ^T y + b = −1 and H ₂ : w ^T y + b = 1 are obtained. The two hyperplanes H ₁ and H ₂ is the distance to and H is parallel to H are substantially equal, between H ₁ and H ₂ because there is no data points between the H ₁ and H ₂ It is something that can not be extended well. In the formula, w and b are parameters. Once the SVM is trained, z selected from the data is classified using w and b.

When non-linearly separable, the “learning machine” maps the input feature vector to a higher dimensional space where the linear hyperplane is potentially located. In this embodiment, the conversion from the nonlinear feature space to the linear high-dimensional space may be performed using a kernel function. Examples of kernels include linear, polynomial, radial basis functions, and sigmoids. For example, a linear kernel may be used for linear SVM processing. Similarly, other kernels may be used for nonlinear SVM processing. In this embodiment, a polynomial kernel is used.

Although a specific system for identifying or classifying marked content, such as images, has been specified, it is desirable to build and evaluate performance, but this is still only one specific embodiment for illustration. Rather, it is noted that the claimed subject matter is not limited in scope to this particular embodiment or approach.

An image database containing 7,560 JPEG images with quality factors of 70-90 was used. One third of the images were a set of substantially random photographs taken at different times and places with different digital cameras. The remaining two thirds were downloaded from the Internet. Each image was trimmed to either 768 x 512 or 512 x 768 size (leaving the central portion). Similarly, for evaluation, the chrominance component of the image is set to zero without changing the luminance coefficient before embedding data.

The above performance evaluation focuses on detection of OutGuess, F5 and MB1 steganography. Codes for these three approaches are publicly available. See the following document: http: // www. outgoings. org / ; http: // wwwrn. inf. tu-dressen. de / ~ westfeld / f5. html; http: // redwood. ucdavis. edu / phil / papers / iwdw03. htm . Since there are quite a number of zero BDCT coefficients in a JPEG image and the number of zero coefficients varies, the data embedding capacity varies from image to image. In general, the ratio of the length of hidden data to the number of non-zero BDCT AC coefficients is used as a measure of the data embedding capacity for JPEG images. For OutGuess, 0.05, 0.1 and 0.2 bpc (bits per non-zero BDCT AC coefficient) were embedded. As a result, 7498, 7452 and 7215 stego images were obtained, respectively. For F5 and MB1, 0.05, 0.1, 0.2, and 0.4 bpc are embedded, and 7560 stego images are obtained. Note that the MB1 embedded step size is equal to 2 in this evaluation.

Half of the images (and associated stego images) were randomly selected to train the SVM classifier and the remaining pairs were used to evaluate the trained classifier. Farid, Shi et al. The above-described method, such as the Fridrich method, and the above-described embodiment were applied to the evaluation detection of the OutGess, F5 and MB schemes. The results shown in Table 2 are arithmetic averages of 20 random tests. Also, as described above, a polynomial kernel was used. The unit here is%; TN represents the true negative rate, TP represents the true positive rate, and AR represents the accuracy.

Similarly, features with a reduced number of dimensions were implemented to investigate the contribution of feature evaluation along various directions. Therefore, one direction feature was implemented at a time. The results shown in Table 4 are arithmetic averages of 20 random tests using a polynomial kernel.

Comparing the tables, it can be seen that the detection rate is improved by combining the directions.

While specific embodiments have been described above, it will be appreciated that the claimed subject matter is not limited in scope to the specific embodiments or implementations. For example, in one embodiment, it may be implemented in hardware and may be implemented to operate, for example, in a device or combination of devices, and in another embodiment may be in software. Similarly, in some embodiments, it may be implemented in firmware or as any combination of hardware, software, and / or firmware, for example. Similarly, in one embodiment, one or more articles such as storage medium (s) may be included, but claimed subject matter is not limited in scope in this respect. For example, these storage media, such as one or more CD-ROMs and / or disks, may have instructions stored thereon, such as executed by a system such as a computer system, computing platform or other system. As a result, an embodiment of the method according to the claimed subject matter may be performed, such as one of the above-described embodiments. As one possible example, a computing platform may include one or more processing units or processors, one or more input / output devices (such as a display, keyboard and / or mouse), and / or one or more memories ( Static random access memory, dynamic random access memory, flash memory, and / or hard drive, etc.). For example, a display may be used to display one or more queries (such as those with correlations) and / or one or more tree representations, again here the claimed subject matter The scope is not limited to this example.

In the foregoing description, various aspects of the claimed subject matter have been described. For purposes of explanation, specific numbers, systems and / or configurations are set forth in order to provide a thorough understanding of claimed subject matter. However, it will be apparent to one skilled in the art having the benefit of this disclosure that the claimed subject matter may be practiced without the specific details set forth above. In other instances, well-known features may be omitted and / or simplified so as not to obscure the claimed subject matter. Although specific features are illustrated and / or described herein, many variations, substitutions, modifications, and / or equivalents will occur to those skilled in the art. Therefore, it is to be understood that the appended claims are intended to cover all such modifications and / or modifications as fall within the spirit of the claimed subject matter.

FIG. 6 is a schematic diagram illustrating one embodiment of a portion of a two-dimensional array of frequency domain coefficients. 1 is a schematic diagram illustrating one embodiment of a technique for generating a frequency domain coefficient array difference. FIG. 1 is a schematic diagram illustrating one embodiment of a technique for generating a frequency domain coefficient array difference. FIG. 1 is a schematic diagram illustrating one embodiment of a technique for generating a frequency domain coefficient array difference. FIG. 1 is a schematic diagram illustrating one embodiment of a technique for generating a frequency domain coefficient array difference. FIG. It is a graph which shows distribution of a coefficient arrangement | sequence difference in one set of image sets. It is a graph which shows distribution of a coefficient arrangement | sequence difference in one set of image sets. FIG. 6 is a schematic diagram illustrating one embodiment of forming a one-step transition probability matrix for characterizing a Markov process. FIG. 6 is a block diagram illustrating one embodiment for generating features.

Claims (33)

Forming at least one coefficient difference array including a plurality of elements, each element generated based at least in part on a difference between the elements of the coefficient array of the frequency domain representation of the at least one training image;
Thresholding the elements of the at least one coefficient difference array to obtain a predetermined range of values to obtain at least one thresholded coefficient difference array; and the at least one threshold process in training image classifier using coefficients difference sequences, wherein at least one training image other than for determining whether the data hidden in at least one image is present, trained hidden data image classification How to get a bowl . The at least one coefficient difference array is formed with respect to at least one of a vertical, horizontal, or diagonal direction;
The method of claim 1. The image classifier includes a distributed processing analyzer for analyzing data based on the statistical variation,
The method of claim 1. The distributed processing analyzer uses SVM processing;
The method of claim 3. The at least one coefficient difference array includes at least three coefficient difference arrays;
The method of claim 1. The at least three coefficient difference arrays include a horizontal coefficient difference array, a vertical coefficient difference array, and a diagonal direction coefficient difference array.
The method of claim 5. The threshold value in the threshold process is variable.
The method of claim 1. The frequency domain representation of the training image includes a DCT block representation;
The method of claim 1. And applying the trained classifier to the image other than the at least one training image,
Further comprising a classifying the images other than the at least one training image based at least in part on the value obtained by application of the trained classifier,
The method of claim 1. An article comprising a computer readable storage medium having stored thereon computer executable instructions comprising:
The instructions are
Instructions for forming at least one coefficient difference array including a plurality of elements, each element generated based at least in part on a difference between elements of a coefficient array of a frequency domain representation of at least one training image;
An instruction for thresholding the elements of the at least one coefficient difference array to obtain the elements within a predetermined range to obtain at least one threshold processed coefficient difference array;
Wherein the image classifier using at least one thresholded coefficient difference sequences in training, said at least one non-training images to determine whether the data hidden in at least one image is present, the training instructions for acquiring has been hidden data image classifier,
Articles containing. The at least one coefficient difference array is formed with respect to at least one of a vertical, horizontal or diagonal direction;
The article of claim 10. The image classifier includes a distributed processing analyzer for analyzing data based on the statistical variation,
The article of claim 10. The distributed processing analyzer uses SVM processing;
The article of claim 12. The at least one coefficient difference array includes at least three coefficient difference arrays;
The article of claim 10. The at least three coefficient difference arrays include a horizontal coefficient difference array, a vertical coefficient difference array, and a diagonal direction coefficient difference array;
The method according to claim 14. The threshold value in the threshold process is variable.
The article of claim 10. The instructions are
Instructions for applying the trained classifier to the at least one training said image other than the image,
Further comprising instructions and classifying the images other than the at least one training image based at least in part on a value obtained by applying the trained classifier,
The article of claim 10. Means for forming at least one coefficient difference array comprising a plurality of elements, each element generated based at least in part on a difference between elements of a coefficient array of a frequency domain representation of at least one training image;
Means for thresholding the elements of the at least one coefficient difference array to obtain the elements within a predetermined range to obtain at least one threshold processed coefficient difference array;
Wherein the image classifier using at least one thresholded coefficient difference sequences in training, said at least one non-training images to determine whether the data hidden in at least one image is present, the training and means for acquiring has been hidden data image classifier. The image classifier includes a distributed processing analyzer for analyzing data based on the statistical variation,
The apparatus according to claim 18. The distributed processing analyzer uses SVM processing.
The apparatus of claim 19. The threshold value in the threshold process is variable.
The apparatus according to claim 18. It means for applying the trained classifier to the image other than the at least one training image,
The apparatus of claim 18, further comprising a means for classifying the image other than the at least one training image based at least in part on a value obtained by application of the trained classifier. The classifying includes classifying the images other than the at least one training image as marked or not marked.
The method of claim 9. By training the image classifier further comprises the frequency-domain coefficients of the training image by applying the amplitude process, to form the frequency domain representation,
The method of claim 1. By training the image classifier includes performing said forming by using a plurality of training images, the threshold processing, and the training,
The method of claim 1. The instruction to classify includes classifying the image other than the at least one training image as marked or not marked.
The article of claim 17. The instruction to train further includes an instruction to input a frequency domain coefficient of the training image by applying amplitude processing to form the frequency domain representation.
The article of claim 10. The instruction to train includes an instruction to perform the formation, the threshold processing, and the training using a plurality of training images.
The article of claim 10. The means for classifying includes means for classifying the images other than the at least one training image as marked or not marked.
The apparatus of claim 22. The means for training further comprises means for inputting the frequency domain coefficients of the at least one training image by applying amplitude processing to form the frequency domain representation.
The apparatus according to claim 18. The means for training includes means for performing the formation, the threshold processing, and the training using a plurality of training images.
The apparatus according to claim 18. An electronic computer comprising at least one processor configured to obtain a trained hidden data image classifier,
The obtaining is
Forming at least one coefficient difference array including a plurality of elements, each element generated based at least in part on a difference between elements of a coefficient array of a frequency domain representation of at least one training image;
Thresholding the elements of the at least one coefficient difference array to make the elements a predetermined range of values to obtain at least one thresholded coefficient difference array;
Training an image classifier with the at least one thresholded coefficient difference array. At least one coefficient difference array including a plurality of elements, each element generated using at least one processor and based at least in part on the differences between the coefficient array elements of the frequency domain representation of the at least one training image Form the
Using the at least one processor to set the element to a predetermined range of values , thresholding the element of the at least one coefficient difference array to obtain at least one threshold processed coefficient difference array And
Using said at least one processor, in training image classifier, wherein said at least one training data hidden in at least one image other than the image exists with the at least one thresholded coefficient difference sequence how to determine whether a method of obtaining the hidden data image classifier trained.

Images