CN109859091A - Image steganography detection method based on Gabor filtering and convolutional neural network - Google Patents
Image steganography detection method based on Gabor filtering and convolutional neural network Download PDFInfo
- Publication number
- CN109859091A CN109859091A CN201811583343.XA CN201811583343A CN109859091A CN 109859091 A CN109859091 A CN 109859091A CN 201811583343 A CN201811583343 A CN 201811583343A CN 109859091 A CN109859091 A CN 109859091A
- Authority
- CN
- China
- Prior art keywords
- image
- steganography
- steganography detection
- gabor
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 148
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 57
- 238000001914 filtration Methods 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 29
- 238000000605 extraction Methods 0.000 claims abstract description 20
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000013139 quantization Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 14
- 238000005070 sampling Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 5
- 238000000034 method Methods 0.000 abstract description 16
- 238000010586 diagram Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 239000000969 carrier Substances 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000001994 activation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000010363 phase shift Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
An image steganography detection method based on Gabor filtering and a convolutional neural network belongs to the technical field of information hiding and is characterized in that: selecting a carrier image and a secret-carrying image to generate a sample image; extracting steganography detection characteristics of the sample image; training the steganography detection characteristics and the class marks of the sample images through an integrated classifier to obtain a steganography detector; and after extracting the steganography detection characteristics of the image to be detected, inputting the steganography detection characteristics into the steganography detector to perform image steganography detection. The method comprises the steps of constructing a plurality of depth convolution neural networks by using a filter for steganography detection feature learning, realizing extraction of diversified learning type steganography detection features, simultaneously extracting constructive type steganography detection features by using a filter coefficient, finally combining the learning type steganography detection features and the constructive type steganography detection features as steganography detection features and performing steganography detection by using an integrated classifier, and remarkably reducing the detection error rate of image self-adaptive steganography.
Description
Technical Field
The invention belongs to the technical field of information hiding, and particularly relates to an image steganography detection method based on Gabor filtering and a convolutional neural network.
Background
The information hiding technique refers to a technique of embedding information into a host signal (such as an image, audio, video, or text file) using insensitivity of human senses and redundancy existing in the signal itself, and detecting or extracting hidden information as necessary. The information hiding technology mainly comprises digital steganography, digital watermarking technology, visible passwords, protocol steganography and the like. Digital steganography is a technique for concealing the existence of secret information by embedding the secret information in redundant data of media such as digital images, audio, video, etc. Digital steganalysis is a reverse technology of digital steganalysis and mainly aims at detecting a steganographic carrier, discovering steganographic communication behaviors, extracting steganographic messages and the like. In a network environment, the universality, the easy acquisition, the sufficient redundant space and other properties of digital images make the digital images become one of the most widely used carrier types in digital steganography, and therefore the digital steganography analysis research is mainly carried out.
The JPEG image is one of the image formats with the widest application range, the mainstream JPEG image self-adaptive steganography algorithm at present comprises UED, J-UNIWARD and the like, and the corresponding steganography detection method comprises a steganography detection method based on high-dimensional features such as DCTR, PHARM, GFR and a component classifier, a steganography detection method based on a deep convolutional neural network and the like. For adaptive steganography detection of JPEG images, the existing detection method can be divided into two frames: one is a frame based on 'steganography detection features + integrated classifiers', the research focuses on the extraction of the steganography detection features, and GFR features are the current mainstream steganography detection features. The other method is a frame based on 'image high-pass filtering + depth convolution neural network', and the research focuses on how to design the depth convolution neural network according to the requirement of steganography detection.
The GFR characteristic is a characteristic with the lowest detection error rate in the current multiple JPEG image self-adaptive steganography detection characteristics, a multi-scale and multi-direction 2D Gabor filter is utilized to filter a decompressed JPEG image, then histogram characteristic extraction is carried out on each filtered residual image according to 64 phases in an 8 x 8DCT block of the JPEG image, histogram characteristic accumulation and combination are carried out according to the symmetry of projection vectors of different phases, then the histogram characteristics of image residual errors generated by the 2D Gabor filter in the symmetry direction are combined, and finally all the extracted histogram characteristics are combined to serve as the final steganography detection characteristics.
The method for detecting the steganography of the deep convolutional neural network based on JPEG phase information, which is proposed by Fridrich et al, mainly comprises three parts: the first part is an image processing layer, and 4 high-pass filters are used for carrying out filtering operation on an image to obtain 4 residual images; the second part is a convolutional neural network layer, which comprises 5 processing modules and is mainly used for carrying out operations such as convolution, data normalization, activation, pooling and the like; and the third part is a classifier layer which comprises a full link layer and a softmax layer and realizes soft decision on the input sample image. The detection method is mainly characterized in that a filtering residual image is divided into 64 sub-images according to JPEG image phase information.
The method for detecting the steganography of the depth convolution neural network based on image residual quantization truncation, which is proposed by Zeng et al, mainly comprises four parts: the first part is an image processing layer, and 25 high-pass filters are used for filtering the image to obtain 25 residual images; the second part is a quantization and truncation layer, 3 quantization operations are carried out on 25 residual images, and truncation thresholds are all 4; the third part is a convolutional neural network layer, and 3 quantization operations correspond to 3 convolutional neural networks; the fourth part is a classifier layer, which comprises a full link network and a softmax layer. The detection method is mainly characterized in that image filtering and residual image quantification and truncation operation are added in a convolutional neural network structure.
However, the above-mentioned prior JPEG image adaptive steganography detection technology has some disadvantages:
1. although GFR is the steganography detection characteristic with the lowest detection error rate, the detection error rate is higher compared with the steganography detection method based on deep learning;
the steganography detection method based on the deep convolutional neural network, which is proposed by Fridrich et al and Zeng et al, has the problem that the types of the filters of the image processing layer are not rich enough;
3. in the prior art, the automatic learning and extraction of the steganography detection features are not realized by utilizing deep learning;
4. the complementarity of the structural steganography detection features and the learning steganography detection features remains to be studied.
Disclosure of Invention
The invention aims to solve the problems and provides an image steganography detection method based on Gabor filtering and a convolutional neural network.
The invention discloses an image steganography detection method based on Gabor filtering and a convolutional neural network, which comprises the steps of selecting a carrier image and a secret-carrying image to generate a sample image;
extracting steganography detection characteristics of the sample image by using a Gabor filter and a depth convolution neural network;
training the steganography detection characteristics and the class marks of the sample images through an integrated classifier to obtain a steganography detector; and after extracting the steganography detection characteristics of the image to be detected, inputting the steganography detection characteristics into the steganography detector to perform image steganography detection.
The image steganography detection method based on Gabor filtering and a convolutional neural network generates a carrier image and a secret-carrying image sample, extracts learning steganography detection characteristics and constructive steganography detection characteristics of the sample image, and combines the learning steganography detection characteristics and the constructive steganography detection characteristics as steganography detection characteristics of the sample image;
combining the steganographic detection features and the class marks of the sample images into a training sample;
training the integrated classifier by using a training sample, and taking the trained classifier as a steganography detector;
extracting learning type steganography detection characteristics and construction type steganography detection characteristics of the image to be detected, and combining the learning type steganography detection characteristics and the construction type steganography detection characteristics to serve as steganography detection characteristics of the image to be detected;
inputting the steganography detection characteristics of the image to be detected into a trained steganography detector for judgment;
and judging that the image to be detected is a carrier image or a secret image according to the output result of the steganography detector.
The image steganography detection method based on Gabor filtering and the convolutional neural network comprises the steps that a carrier image class mark is 1, and a secret-carrying image class mark is-1.
The invention discloses an image steganography detection method based on Gabor filtering and a convolutional neural network, which comprises the following steps of: firstly, filtering a sample image by using a multi-scale and multi-directional 2D Gabor filter, and then training a deep convolutional neural network by using filtering images obtained by using the 2D Gabor filters with the same scale parameters to obtain a plurality of trained deep convolutional neural networks; during learning type feature extraction, multi-scale and multidirectional 2D Gabor filtering is carried out on a sample image, then filtering images with different scales are input into corresponding trained deep convolutional neural networks to obtain a plurality of corresponding learning type steganography detection features, and finally the obtained learning type steganography detection features are combined to serve as the final learning type steganography detection feature.
The invention discloses an image steganography detection method based on Gabor filtering and a convolutional neural network, which comprises the following steps of: firstly, decompressing an image, and then performing high-pass filtering operation on the decompressed image by using a multi-scale and multidirectional 2DGabor filter to obtain a residual image; secondly, carrying out multiple quantization operations on each residual image by utilizing multiple quantization step lengths, and further respectively carrying out histogram feature extraction on the multiple quantized images and accumulating and combining the histogram features to be used as histogram features corresponding to one residual image; and then combining the histogram features of the residual image generated by the 2D Gabor filters in the symmetrical directions, and finally combining all the combined histogram features to serve as the final structural steganography detection feature.
According to the image steganography detection method based on Gabor filtering and the convolutional neural network, the image format is JPEG.
The invention discloses an image steganography detection method based on Gabor filtering and a convolutional neural network, which comprises the following steps of:
step 1: generating 2D Gabor filters with different parameters; a 2D Gabor function shown below is used to generate a 2D Gabor filter,
wherein x 'xcos θ + ysin θ, y' xsin θ + ycos θ, σ is 0.56 λ, and γ is 0.5;
step 2: generating a residual image; carrying out high-pass filtering operation on the input training sample image by using the 2D Gabor filter generated in the last step to obtain a filtering residual image;
step 3: training a deep convolutional neural network; respectively training a deep convolutional neural network by using the filtered residual images;
step 4: extracting learning type steganography detection characteristics; and performing convolution operation on the input image and a 2D Gabor filter, then inputting the obtained residual images into corresponding deep convolutional neural networks, taking the characteristics output by each deep convolutional neural network, and finally combining the obtained multiple learning type steganography detection characteristics into a final learning type steganography detection characteristic.
The invention discloses an image steganography detection method based on Gabor filtering and a convolutional neural network, wherein the 2D Gabor filter is generated by the following steps: (1) generating sampling points; assuming that the 2D Gabor filter size is MxN, the value range of x isy has a value range ofGenerating a sampling point (x, y) by step 1; (2) determining filter parameters; determining the sum of the parameters sigma, theta in a 2D Gabor functionThe value of the parameter λ is calculated, γ is 0.5; (3) generating a 2DGabor filter; calculating a function value at a sampling point according to the 2D Gabor function expression and the parameter value to obtain a corresponding 2DGabor filter; (4) normalizing the 2D Gabor filter; the 2D Gabor filter is subtracted from the mean of all its elements to obtain a zero mean high pass filter.
The image steganography detection method based on Gabor filtering and convolutional neural network utilizes a multi-scale and multidirectional 2D Gabor filter to carry out image filtering, then constructing a plurality of deep convolution neural networks to perform steganography detection feature learning, realizing the extraction of diversified learning type steganography detection features, meanwhile, the method also utilizes the 2D Gabor filter coefficient of the JPEG image to extract the structural steganography detection characteristic, finally combines the learning type steganography detection characteristic and the structural steganography detection characteristic as the steganography detection characteristic and utilizes an integrated classifier to perform steganography detection, the steganography detection method significantly reduces the detection error rate of the adaptive steganography of the JPEG image, for J-UNIWARD steganography with the quality factor of 75 and the embedding ratio of 0.3, the detection error rate of the steganography detection method is reduced by 12 percent compared with the detection error rate of GFR and is reduced by 6 percent compared with methods such as Jessia and the like.
Drawings
FIG. 1 is a schematic diagram of a steganographic detection feature construction process according to the present invention;
FIG. 2 is a schematic diagram of learning-based steganography detection feature extraction according to the present invention;
FIG. 3 is a schematic diagram of the feature extraction for structural steganography detection according to the present invention;
FIG. 4 is a schematic diagram of a classifier training process according to the present invention;
FIG. 5 is a schematic view of a process flow of an image to be detected according to the present invention;
FIG. 6 is a schematic diagram of the deep convolutional neural network structure according to the present invention;
FIG. 7 is a merged schematic diagram of 64 (T +1) -dimensional histogram features when the 2D Gabor filter orientation parameter θ is 0 or π/2 according to the present invention;
FIG. 8 is a diagram illustrating a combination of 64 (T +1) -dimensional histogram features when the 2D Gabor filter direction parameter θ takes other values except 0 and π/2.
Detailed Description
The following describes in detail an image steganography detection method based on a filter and a deep convolutional neural network according to the present invention with reference to the accompanying drawings and embodiments.
Example one
The invention discloses an image steganography detection method based on Gabor filtering and a convolutional neural network, which comprises the following steps of: generating a carrier image and a secret-carrying image sample, extracting learning type steganography detection characteristics and structural type steganography detection characteristics of the sample image shown in fig. 1, and combining the learning type steganography detection characteristics and the structural type steganography detection characteristics as steganography detection characteristics of the sample image; as shown in fig. 2, the step of extracting the learning-type steganography detection feature includes: firstly, filtering JPEG images by utilizing a multi-scale and multi-direction 2DGabor filter, and then respectively training a deep convolutional neural network by utilizing filtering images obtained by the 2D Gabor filter with the same scale parameters to obtain a plurality of trained deep convolutional neural networks; and secondly, carrying out multi-scale and multi-direction 2D Gabor filtering on the sample image, inputting the filtered sample image into a corresponding trained deep convolutional neural network to obtain a plurality of learning type steganography detection features, and finally combining the obtained plurality of learning type steganography detection features into a final learning type steganography detection feature.
As shown in fig. 3, the step of extracting the structural steganography detection feature includes: firstly, decompressing a JPEG image, and then performing high-pass filtering operation on the decompressed image by using a multi-scale and multidirectional 2D Gabor filter to obtain a residual image; secondly, carrying out multiple quantization operations on each residual image by using multiple quantization step lengths, respectively carrying out histogram feature extraction on the multiple quantized images, accumulating and combining the histogram features to be used as histogram features corresponding to one residual image, then combining the histogram features of the residual image generated by a 2D Gabor filter in the symmetrical direction, and finally combining all the combined histogram features to be used as a final structural steganography detection feature.
As shown in fig. 4, the steganographic detection features and the class labels of the sample image are combined into a training sample; the carrier image class is marked as 1, and the secret-carrying image class is marked as-1;
training the integrated classifier by using a training sample, and taking the trained classifier as a steganography detector;
as shown in fig. 5, the learning steganography detection feature and the structural steganography detection feature of the image to be detected are extracted, and the combination of the learning steganography detection feature and the structural steganography detection feature is used as the steganography detection feature of the image to be detected;
inputting the steganography detection characteristics of the image to be detected into a trained steganography detector for judgment;
and judging that the image to be detected is a carrier image or a secret image according to the output result of the steganography detector.
Example two
On the basis of the first embodiment, the generating of residual images by using different sets of 2D Gabor filters, and then training a plurality of deep convolutional neural networks by using the residual images and implementing learning-type feature extraction comprises the following steps:
step 1: four groups of 2D Gabor filters for generating different parameters
A 2D Gabor filter is generated using a 2D Gabor function shown in the following equation,
where x '═ xcos θ + ysin θ, y' ═ xsin θ + ycos θ, σ ═ 0.56 λ, and γ ═ 0.5.
The 2D Gabor filter generation steps are as follows: (1) and generating sampling points. Assuming that the 2D Gabor filter size is MxN, the value range of x isy has a value range ofGenerating a sampling point (x, y) by step 1; (2) the filter parameters are determined. Determining the sum of the parameters sigma, theta in a 2D Gabor functionThe value of the parameter λ is calculated, γ is 0.5; (3) a 2D Gabor filter is generated. According toCalculating a function value at a sampling point by using the 2D Gabor function expression and the parameter value to obtain a corresponding 2D Gabor filter; (4) normalizing the 2D Gabor filter; the 2D Gabor filter is subtracted from the mean of all its elements to obtain a zero mean high pass filter.
The scale parameter of the 2D Gabor filter is 8 multiplied by 8, the scale parameter sigma is respectively 0.75, 1, 1.25 and 1.5, the direction parameter theta is respectively {0, pi/4, 2 pi/4, 3 pi/4 } corresponding to each scale parameter sigma, and the phase parameter theta is respectivelyAnd respectively taking 0 and pi/2, and forming a filter group by 8 2D Gabor filters with the same scale parameter to obtain four groups of 2D Gabor filters.
Step 2: generating four sets of residual images
As shown in fig. 2, four sets of 2D Gabor filters generated in the previous step are used to perform a high-pass filtering operation on the input training sample image, so as to obtain four sets of filtered residual images.
Step 3: training four deep convolutional neural networks
As shown in fig. 2, four deep convolutional neural networks are respectively trained by using four groups of filtered residual images, and the structure of each deep convolutional neural network is shown in fig. 6. The deep convolutional neural network shown in fig. 6 can be structurally divided into 5 groups, each group includes several layers, and the convolutional filter in each layer is represented by the form: the number of filters x (filter height x filter width x number of input feature maps) is represented as: feature map number x (feature map height x feature map width); ABS means absolute value layer, i.e. taking absolute value of the filtered image; batch Normalization (BN) is a Batch Normalization layer, namely, a characteristic diagram is subjected to Batch Normalization operation, so that the mean value is 0, and the variance is 1; TanH and ReLU are activation functions, and the TanH activation functions in Group1 and Group2 can limit data within a smaller value range; avg Pooling represents the average Pooling layer; phase Split means that sub-images of a feature image obtained from a JPEG image are sampled in 64 phases in 8 × 8 blocks, and 64 sub-images can be obtained from one feature image.
When the deep convolutional neural network is trained, a small batch of random gradient descent algorithm is used for network training, and the loss function is a cross entropy function. In the random gradient descent algorithm, the momentum value is set to be 0.9, the parameter learning rate is initially set to be 0.001, the learning rate is reduced by 25% every 20 generations in the training process, and the total iteration number of the training algorithm is 240 generations; each batch of training samples comprises 40 images, wherein 20 carriers and 20 carriers are densely loaded; all convolution filters in the convolutional neural network are initialized by Gaussian distribution with the mean value of 0 and the standard deviation of 0.01, bias parameters in the convolutional layers are all fixed to be 0, the learning rate of the bias parameters in the full link layer is twice of the learning rate of other parameters of the network, other parameters in the full link layer are initialized by adopting an Xavier method, the L2 norm weight attenuation coefficient in the full link layer is set to be 0.01, and other layers have no weight attenuation.
Step 4: extracting learning-based steganography detection features
After the four deep convolutional neural networks are trained, learning steganography detection feature extraction can be performed on the images by using the networks. The method comprises the following steps: carrying out convolution operation on the input image and four groups of 2D Gabor filters respectively, then inputting the four groups of residual images into four corresponding deep convolution neural networks respectively, taking 512-dimensional features output by Group5 in each deep convolution neural network as learning-type features, and obtaining 2048-dimensional learning-type steganography detection features by the four deep convolution neural networks.
EXAMPLE III
As shown in fig. 3, different sets of 2D Gabor filters are used to generate filtered residual images, then the residual images are quantized with different step sizes, and finally histogram feature extraction is performed on the quantized residual images. The method comprises the following specific steps:
step 1: decompressing the JPEG image to a space domain without rounding;
step 2: generating a 2D Gabor filter bank, wherein the scale parameter sigma takes 0.75, 1, 1.25 and 1.5, the direction parameter sigma corresponding to each scale takes {0, pi/16, 2 pi/16, …,15 pi/16 } respectively, and the phase deviation parameter sigma takesRespectively taking 0 and pi/2 to obtain four groups of 2DGabor filters, wherein the number of the filters in each group is 32, and 128 filters are obtained in total;
step 3: decompressing JPEG image with each 2D Gabor filter G in filter bankσ,θPerforming convolution to obtain a filtering residual image Uσ,θ；
Step 4: for filtering residual error image Uσ,θCarrying out quantization of a plurality of different step sizes, then carrying out histogram feature extraction on the filtered residual image of each quantization step size according to the following formula,
wherein,representing the filtered residual image, Q the quantization step, T the truncation threshold and taken to be 4, QTRepresenting a quantization function with a quantization center of 0,1,2, …, T, (a, b) representing the position in an 8 x 8 image block.
For the quantization step q, q' is calculated according to the following formula,
wherein QF represents a JPEG image quality factor, and then the filtering residual image is quantized by taking 0.5q ', q ' and 1.5q ' as quantization step sizes respectively.
Step 5: according to the histogram feature extraction formula, 64 (T +1) -dimensional histogram features can be extracted from each residual image, when the corresponding 2D Gabor filter direction parameter theta is 0 and pi/2, according to the symmetry relation shown in FIG. 7, the histogram features at the same positions are merged to finally obtain a 25 x (T +1) -dimensional histogram feature, and when the direction parameter theta takes other values, according to the symmetry relation shown in FIG. 8, a 34 x (T +1) -dimensional histogram feature is merged.
Step 6: merging histogram features extracted by a 2D Gabor filter with the same scale and phase shift parameters and symmetrical direction;
step 7: combining the histogram features extracted from all the filtered images with the same quantization step size to obtain a 2 × (4 × (2 × 25+ (16/2-1) × 34) × (4+1)) dimensional histogram feature;
step 8: finally, histogram features extracted from different quantization step size filtered residual images are accumulated to form a final steganography detection feature, and the feature dimension is still 2 × (4 × (2 × 25+ (16/2-1) × 34) × (4+1)) -11520 dimensions.
Claims (8)
1. The image steganography detection method based on Gabor filtering and the convolutional neural network is characterized by comprising the following steps:
selecting a carrier image and a secret-carrying image to generate a sample image;
extracting steganography detection characteristics of the sample image by using a Gabor filter and a depth convolution neural network;
training the steganography detection characteristics and the class marks of the sample images through an integrated classifier to obtain a steganography detector;
and after extracting the steganography detection characteristics of the image to be detected, inputting the steganography detection characteristics into the steganography detector to perform image steganography detection.
2. The image steganography detection method based on Gabor filtering and convolutional neural network of claim 1, wherein:
generating a carrier image and a secret-carrying image sample, extracting learning type steganography detection characteristics and structural type steganography detection characteristics of the sample image, and combining the learning type steganography detection characteristics and the structural type steganography detection characteristics as steganography detection characteristics of the sample image;
combining the steganographic detection features and the class marks of the sample images into a training sample;
training the integrated classifier by using a training sample, and taking the trained classifier as a steganography detector;
extracting learning type steganography detection characteristics and construction type steganography detection characteristics of the image to be detected, and combining the learning type steganography detection characteristics and the construction type steganography detection characteristics to serve as steganography detection characteristics of the image to be detected;
inputting the steganography detection characteristics of the image to be detected into a trained steganography detector for judgment;
and judging that the image to be detected is a carrier image or a secret image according to the output result of the steganography detector.
3. The image steganography detection method based on Gabor filtering and convolutional neural network of claim 2, wherein: the class mark comprises a carrier image class mark 1, and a secret-carrying image class mark-1.
4. The image steganography detection method based on Gabor filtering and convolutional neural network of claim 2 or 3, wherein: the extraction step of the learning type steganography detection features comprises the following steps: firstly, filtering a sample image by using a multi-scale and multi-direction 2D Gabor filter, and then training a deep convolutional neural network by using filtering images obtained by using 2DGabor filters with the same scale parameters to obtain a plurality of trained deep convolutional neural networks; during learning type feature extraction, multi-scale and multi-direction 2DGabor filtering is carried out on a sample image, then filtering images with different scales are input into a corresponding trained deep convolutional neural network to obtain a plurality of learning type steganography detection features, and finally the obtained learning type steganography detection features are combined into a final learning type steganography detection feature.
5. The image steganography detection method based on Gabor filtering and convolutional neural network of claim 4, wherein: the extraction step of the structural steganography detection features comprises the following steps: firstly, decompressing an image, and then performing high-pass filtering operation on the decompressed image by using a multi-scale and multidirectional 2D Gabor filter to obtain a residual image; secondly, carrying out multiple quantization operations on each residual image by utilizing multiple quantization step lengths, further respectively carrying out histogram feature extraction on the multiple quantized images, accumulating and combining the histogram features to be used as the histogram features corresponding to one residual image, then combining the histogram features of the residual image generated by a 2D Gabor filter in the symmetrical direction, and finally combining all the combined histogram features to be used as the final structural steganography detection feature.
6. The image steganography detection method based on Gabor filtering and convolutional neural network of claim 5, wherein: the image format is JPEG.
7. The image steganography detection method based on Gabor filtering and convolutional neural network of claim 6, wherein: the learning type steganography detection feature extraction comprises the following steps:
step 1: generating 2D Gabor filters with different parameters; a 2D Gabor function shown below is used to generate a 2D Gabor filter,
where x '═ xcos θ + ysin θ, y' ═ xsin θ + ycos θ, σ ═ 0.56 λ, and γ ═ 0.5.
Step 2: generating a residual image; carrying out high-pass filtering operation on the input training sample image by using the 2D Gabor filter generated in the last step to obtain a filtering residual image;
step 3: training a deep convolutional neural network; respectively training a deep convolutional neural network by using the filtered residual images;
step 4: extracting learning type steganography detection characteristics; and performing convolution operation on the input image and a 2D Gabor filter, then inputting the obtained residual images into corresponding deep convolutional neural networks, taking the characteristics output by each deep convolutional neural network, and finally combining the obtained multiple learning type steganography detection characteristics into a final learning type steganography detection characteristic.
8. The image steganography detection method based on Gabor filtering and convolutional neural network of claim 7, wherein: the 2D Gabor filter is generated by the following steps: (1) generating sampling points; assuming that the 2D Gabor filter size is MxN, the value range of x isy has a value range ofGenerating a sampling point (x, y) by step 1; (2) determining filter parameters; determining the sum of the parameters sigma, theta in a 2DGabor functionThe value of the parameter λ is calculated, γ is 0.5; (3) generating a 2D Gabor filter; calculating a function value at a sampling point according to the 2D Gabor function expression and the parameter value to obtain a corresponding 2D Gabor filter; (4) normalizing the 2D Gabor filter; the 2D Gabor filter is subtracted from the mean of all its elements to obtain a zero mean high pass filter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811583343.XA CN109859091B (en) | 2018-12-24 | 2018-12-24 | Image steganography detection method based on Gabor filtering and convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811583343.XA CN109859091B (en) | 2018-12-24 | 2018-12-24 | Image steganography detection method based on Gabor filtering and convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109859091A true CN109859091A (en) | 2019-06-07 |
| CN109859091B CN109859091B (en) | 2023-05-16 |
Family
ID=66892030
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811583343.XA Active CN109859091B (en) | 2018-12-24 | 2018-12-24 | Image steganography detection method based on Gabor filtering and convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109859091B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245621A (en) * | 2019-06-17 | 2019-09-17 | 深圳Tcl新技术有限公司 | Face identification device and image processing method, Feature Selection Model, storage medium |
| CN110490265A (en) * | 2019-08-23 | 2019-11-22 | 安徽大学 | A kind of image latent writing analysis method based on two-way convolution sum Fusion Features |
| CN110619594A (en) * | 2019-09-16 | 2019-12-27 | 中山大学 | Halftone image steganalysis method |
| CN111028308A (en) * | 2019-11-19 | 2020-04-17 | 珠海涵辰科技有限公司 | Steganography and reading method for information in image |
| CN111507884A (en) * | 2020-04-19 | 2020-08-07 | 衡阳师范学院 | Self-adaptive image steganalysis method and system based on deep convolutional neural network |
| CN111859897A (en) * | 2019-10-16 | 2020-10-30 | 沈阳工业大学 | Text steganalysis method based on dynamic routing capsule network |
| CN115147501A (en) * | 2022-09-05 | 2022-10-04 | 深圳市明源云科技有限公司 | Picture decompression method and device, terminal device and storage medium |
| CN115731089A (en) * | 2022-12-16 | 2023-03-03 | 中国人民解放军61660部队 | Component energy-based double-task image steganography method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101151622A (en) * | 2005-01-26 | 2008-03-26 | 新泽西理工学院 | System and method for steganalysis |
| US20090220076A1 (en) * | 2008-02-28 | 2009-09-03 | Fujitsu Limited | Image decrypting apparatus, image encrypting apparatus, and image decrypting method |
| CN108961137A (en) * | 2018-07-12 | 2018-12-07 | 中山大学 | A kind of image latent writing analysis method and system based on convolutional neural networks |
-
2018
- 2018-12-24 CN CN201811583343.XA patent/CN109859091B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101151622A (en) * | 2005-01-26 | 2008-03-26 | 新泽西理工学院 | System and method for steganalysis |
| US20090220076A1 (en) * | 2008-02-28 | 2009-09-03 | Fujitsu Limited | Image decrypting apparatus, image encrypting apparatus, and image decrypting method |
| CN108961137A (en) * | 2018-07-12 | 2018-12-07 | 中山大学 | A kind of image latent writing analysis method and system based on convolutional neural networks |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245621A (en) * | 2019-06-17 | 2019-09-17 | 深圳Tcl新技术有限公司 | Face identification device and image processing method, Feature Selection Model, storage medium |
| CN110245621B (en) * | 2019-06-17 | 2023-10-17 | 深圳Tcl新技术有限公司 | Face recognition device, image processing method, feature extraction model, and storage medium |
| CN110490265A (en) * | 2019-08-23 | 2019-11-22 | 安徽大学 | A kind of image latent writing analysis method based on two-way convolution sum Fusion Features |
| CN110490265B (en) * | 2019-08-23 | 2022-04-15 | 安徽大学 | Image steganalysis method based on double-path convolution and feature fusion |
| CN110619594A (en) * | 2019-09-16 | 2019-12-27 | 中山大学 | Halftone image steganalysis method |
| CN111859897A (en) * | 2019-10-16 | 2020-10-30 | 沈阳工业大学 | Text steganalysis method based on dynamic routing capsule network |
| CN111028308A (en) * | 2019-11-19 | 2020-04-17 | 珠海涵辰科技有限公司 | Steganography and reading method for information in image |
| CN111028308B (en) * | 2019-11-19 | 2022-11-04 | 珠海涵辰科技有限公司 | Steganography and reading method for information in image |
| CN111507884A (en) * | 2020-04-19 | 2020-08-07 | 衡阳师范学院 | Self-adaptive image steganalysis method and system based on deep convolutional neural network |
| CN115147501A (en) * | 2022-09-05 | 2022-10-04 | 深圳市明源云科技有限公司 | Picture decompression method and device, terminal device and storage medium |
| CN115731089A (en) * | 2022-12-16 | 2023-03-03 | 中国人民解放军61660部队 | Component energy-based double-task image steganography method |
| CN115731089B (en) * | 2022-12-16 | 2024-02-23 | 中国人民解放军61660部队 | Dual-task image steganography method based on component energy |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109859091B (en) | 2023-05-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109859091B (en) | Image steganography detection method based on Gabor filtering and convolutional neural network | |
| CN110211045B (en) | Super-resolution face image reconstruction method based on SRGAN network | |
| Babu et al. | Statistical features based optimized technique for copy move forgery detection | |
| CN105657402B (en) | A kind of depth map restoration methods | |
| CN109035149B (en) | License plate image motion blur removing method based on deep learning | |
| CN113222800B (en) | Robust image watermark embedding and extracting method and system based on deep learning | |
| Kang et al. | Robust median filtering forensics using an autoregressive model | |
| CN115063573B (en) | Multi-scale target detection method based on attention mechanism | |
| CN112907598B (en) | Method for detecting falsification of document and certificate images based on attention CNN | |
| CN108961137A (en) | A kind of image latent writing analysis method and system based on convolutional neural networks | |
| CN102930518B (en) | Improved sparse representation based image super-resolution method | |
| CN107092884B (en) | Rapid coarse-fine cascade pedestrian detection method | |
| CN111681188B (en) | Image deblurring method based on combination of image pixel prior and image gradient prior | |
| CN113344110B (en) | Fuzzy image classification method based on super-resolution reconstruction | |
| CN117274059A (en) | Low-resolution image reconstruction method and system based on image coding-decoding | |
| CN109543672A (en) | Object detecting method based on dense characteristic pyramid network | |
| CN106886763A (en) | The system and method for real-time detection face | |
| CN118115729B (en) | Image fake region identification method and system with multi-level and multi-scale feature interaction | |
| CN112862655B (en) | JPEG image steganalysis method based on channel space attention mechanism | |
| CN116385281A (en) | Remote sensing image denoising method based on real noise model and generated countermeasure network | |
| CN117726954B (en) | Sea-land segmentation method and system for remote sensing image | |
| Liang et al. | Image resampling detection based on convolutional neural network | |
| Yu et al. | A multi-task learning CNN for image steganalysis | |
| CN113392728B (en) | Target detection method based on SSA sharpening attention mechanism | |
| Rana et al. | MSRD-CNN: Multi-scale residual deep CNN for general-purpose image manipulation detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |